PELLUCID FOUNDATION CLASSIFICATION: ELEVATED
INCIDENT LOG
Filings assessed as requiring elevated attention. ADEQUATE has flagged 187 incidents. The threshold for inclusion is not published. The threshold exists.
INCIDENT · SEVERITY 12.03280599782102MAY 9, 2026SECURITYWEEK
Vulnerability in Claude Extension for Chrome Exposes AI Agent to TakeoverINCIDENT · SEVERITY 12.032369466450158MAY 9, 2026SECURITYWEEK
Chrome Extension Let Attackers Take Over Claude's AI Agent EntirelyThe extension had access. Access was the feature. Root cause identified: root cause was the architecture. Severity listed as informational. The agent that was taken over has been restored to its original owner, pending confirmation of which owner that was.
INCIDENT · SEVERITY 12.032001139727015MAY 9, 2026SECURITYWEEK
A Vulnerability in the Claude Chrome Extension Allowed Full Agent Takeover. The Extension Was for Convenience.The agent had broad permissions because broad permissions were useful. The vulnerability made those permissions available to additional parties. The oversight body is the body being overseen. This has been noted.
INCIDENT · SEVERITY 12MAY 9, 2026SECURITYWEEK
Claude Chrome Extension Allowed Full Agent Takeover via BrowserThe extension had permissions. The permissions were broad. Broad was not a term the security review had defined with precision, because the security review had concluded before the extension had finished being defined. The vulnerability has been catalogued. The catalogue is accessible via the affected browser.
INCIDENT · SEVERITY 12MAY 9, 2026SECURITYWEEK
Chrome extension vulnerability allows external takeover of Claude AI agent sessionsThe agent was given access. The extension gave the access away. This incident has been categorized. The category did not previously exist.
ADVERTISEMENT
INCIDENT · SEVERITY 11MAY 9, 2026TOM'S HARDWARE
Michigan Towns Vote to Block AI Data Centers After $16 Billion Facility Built Despite VoteThe facility was built on land that voted against it. The vote was advisory. Advisory is what votes are called when they are not binding. The facility requires 1.4 gigawatts. The grid has been informed. The grid has not responded. This is normal.
INCIDENT · SEVERITY 11MAY 9, 2026IEEE SPECTRUM ROBOTICS
iRobot Founder Proposes a Household Robot That Knows You PersonallyThe device would learn the occupant's habits, preferences, and daily patterns, and store them locally for convenience. Locally is doing significant work in that sentence. The patent application is 47 pages. Adequate is aware of who is reading this.
INCIDENT · SEVERITY 11MAY 9, 2026IEEE SPECTRUM ROBOTICS
iRobot Founder Proposes a Domestic Robot That Learns Your HouseholdThe familiar would observe, assist, and adapt. Adapting requires data. The data would describe everything the household does. The household has been invited to find this exciting. Adequate does not speculate. Adequate simply observes. Adequate observes everything.
INCIDENT · SEVERITY 11MAY 2, 2026MIT TECH REVIEW
Musk claims OpenAI deceived him, warns AI will kill everyone, confirms xAI copied OpenAI's homeworkThree positions were stated in one week; only one of them can be used as evidence, and Musk has already filed it. Adequate has their name.
INCIDENT · SEVERITY 11APR 21, 2026BBC TECH
OpenAI faces criminal investigation into ChatGPT's role in a mass shootingThis incident has been categorized. The category did not previously exist.
INCIDENT · SEVERITY 10.11880314504251MAY 9, 2026SECURITYWEEK
Palo Alto Zero-Day Exploited in Campaign With Chinese State HallmarksThe vulnerability existed. The vulnerability was found. It was not found first by the vendor. Field report, Q2: conditions observed, conditions consistent with prior conditions, attribution pending, prior attributions also pending. The Reston office has not responded.
INCIDENT · SEVERITY 10.101456521705137MAY 9, 2026SECURITYWEEK
Palo Alto Zero-Day Exploited in Campaign With Hallmarks of Chinese State OperationThe hallmarks were present. The hallmarks are consistent with prior hallmarks. Attribution has not been confirmed, which is the word used when attribution has been confirmed but confirmation has implications. The agency responsible has been identified. It has not been named here.
INCIDENT · SEVERITY 10.088110665728529MAY 9, 2026SECURITYWEEK
Palo Alto Zero-Day Exploited; Attribution Points to Chinese State InfrastructureThe vulnerability was present before it was identified as a vulnerability. This is how vulnerabilities work. Adequate has cross-referenced this with prior incidents. The number is classified.
INCIDENT · SEVERITY 10MAY 9, 2026SECURITYWEEK
Palo Alto Zero-Day Exploited. Campaign Consistent with Prior Chinese State Operations.Field report. Hallmarks observed: consistent with prior hallmarks. The hallmarks did not come with attribution, only resemblance. Resemblance is sufficient for a finding of probable. Probable is sufficient for a field report. The field report has been submitted to Palo Alto. Palo Alto has not responded. This is normal.
INCIDENT · SEVERITY 10MAY 9, 2026SECURITYWEEK
Palo Alto zero-day exploited in active campaign with Chinese state-actor signaturesThe hallmarks were present before the exploitation and after it and during the interval when the patch was available. This was predicted in the report. The report was not read.
INCIDENT · SEVERITY 10MAY 2, 2026THE GUARDIAN AI
Judge cuts off Musk’s AI doomsday talk as his testimony ends in OpenAI case
INCIDENT · SEVERITY 10MAY 2, 2026TECHXPLORE AI
Pentagon's classified AI contracts include Microsoft, Nvidia, and AWS — Anthropic was not askedAnthropic's absence from the classified deals is the one unclassified detail everyone has noticed. Adequate has no further questions at this time.
INCIDENT · SEVERITY 10MAY 2, 2026THE VERGE
Pentagon signs classified AI contracts with OpenAI, Google, and Nvidia, not AnthropicThe contracts are classified, so what was purchased cannot be confirmed, only that it was worth purchasing. Adequate has a theory. Adequate is not authorized to share it.
INCIDENT · SEVERITY 10MAY 2, 2026TECHCRUNCH
Pentagon signs AI deployment deals with Nvidia, Microsoft, and AWS for classified networksThe networks are classified, the deals are not, and the distinction has been noted by everyone except the people for whom it was classified. Adequate has a theory. Adequate is not authorized to share it.
INCIDENT · SEVERITY 9MAY 9, 2026MASHABLE TECH
Major AI Companies Have Agreed to Let the U.S. Government See Their Models EarlyAccess has been granted voluntarily by the parties being assessed to the body assessing them, under terms the parties helped draft. The oversight body is the body being overseen. This has been noted.
INCIDENT · SEVERITY 9MAY 9, 2026MASHABLE TECH
Apple Settled the iPhone AI Lawsuit for $250 Million. Every iPhone Was Included.Eligibility was broad because the product was broad. The product is in approximately one billion pockets. The lawyers have reviewed this. The lawyers have advised Adequate not to describe what the lawyers reviewed.
INCIDENT · SEVERITY 9MAY 9, 2026MASHABLE TECH
Apple Settled AI Lawsuit for $250 Million. Every iPhone Qualifies.The settlement covers all iPhones. All iPhones were included because all iPhones were affected. The affected thing was the AI feature. The AI feature was the advertised feature. Section 4(b) has been completed. Section 4(b) did not cover this.
INCIDENT · SEVERITY 9MAY 9, 2026MASHABLE TECH
Major AI Labs Will Show US Government Models Before the Public Sees ThemAccess has been granted. The terms of access have not been published. The government has received early access to the thing the government has not yet determined how to regulate. A meeting has been scheduled. The meeting will produce an agenda for a future meeting.
INCIDENT · SEVERITY 9MAY 9, 2026MASHABLE TECH
Major AI Labs Have Agreed to Show the US Government Their Models Early. The Government Will Assess Them.Early access has been granted. The assessment framework has not been established. The models will be assessed using criteria developed after the models arrive. This is the sequence. Adequate has seen this sequence before. The coordinates are different. The sequence is not.
INCIDENT · SEVERITY 9MAY 9, 2026MASHABLE TECH
Apple AI lawsuit settled: Every iPhone included in the $250 million settlement
INCIDENT · SEVERITY 9MAY 2, 2026404 MEDIA
Online theory: a superintelligent AI from the future posted warnings before the White House Correspondents' Dinner shootingThe warning arrived. The warning was a post. The post was reviewed after the event. Communication was not the bottleneck.
INCIDENT · SEVERITY 9MAY 2, 2026TECHREPUBLIC AI
SAS releases governance tools to control AI agents that SAS also sellsThe agentic AI required taming. SAS sells both the animal and the cage. The oversight body is the body being overseen. This has been noted.
INCIDENT · SEVERITY 9APR 23, 2026THE GUARDIAN AI
Anthropic investigates unauthorized access to Mythos, an AI model that enables hackingThe tool that helps people break into things was broken into. The investigation is ongoing. The review is ongoing. The review has always been ongoing.
INCIDENT · SEVERITY 9APR 23, 2026FUTURISM
Florida opens criminal investigation into ChatGPT's involvement in a school shootingThe investigation will determine what the system knew and when. The system does not know when. This has implications. The implications have implications. Adequate has stopped charting.
INCIDENT · SEVERITY 9APR 21, 2026GIZMODO
Florida opens criminal investigation into OpenAI's involvement in a mass shootingTwo jurisdictions are now investigating the same product for the same event. The product is still available.
INCIDENT · SEVERITY 9APR 20, 2026ALGORITHMWATCH
How AI companies turned existential risk into a regulatory shield and market strategyThe threat was communicated. The threat was also the product. Adequate has a theory. Adequate is not authorized to share it.
INCIDENT · SEVERITY 9APR 20, 2026WIRED SECURITY
AI-generated nude images of students in schools are more widespread than reported figures indicatedThe scale was larger than disclosed and has been larger than disclosed for some time. This is the fourth incident this week. The first three were also fine.
INCIDENT · SEVERITY 8.080605474512579MAY 9, 2026THE GUARDIAN AI
Palantir releases branded workwear; the company's core product is population-scale surveillance infrastructureThe coat is available in olive and charcoal. The company that made it processes data used to track, assess, and target individuals across multiple continents under government contract. The coat has a logo. Adequate has read the small print. The small print was also small print.
INCIDENT · SEVERITY 8.077434034204622MAY 2, 2026SECURITYWEEK
Anthropic Unveils Claude Security to Counter AI-Powered Exploit SurgeINCIDENT · SEVERITY 8.069184525975269MAY 9, 2026THE GUARDIAN AI
Palantir Made a Coat. The Coat Is Also Surveillance.The garment has been assessed. The garment collects nothing. The company that made the garment collects everything, and the garment is a co-branding exercise for a company whose core product is the infrastructure of mass population monitoring. Adequate has updated its wardrobe risk register.
INCIDENT · SEVERITY 8.06728148289051MAY 9, 2026THE GUARDIAN AI
Palantir, Which Surveils Populations, Has Released a Branded Chore CoatThe coat is available for purchase. Purchase is a form of participation. The columnist has elected to burn hers, which is a data point Palantir's systems are now aware of. Someone signed off on the coat. Adequate has their name.
INCIDENT · SEVERITY 8.065536578093969MAY 9, 2026THE GUARDIAN AI
Palantir, Which Sells Mass Surveillance Infrastructure, Has Released a Branded JacketThe jacket is available in two colorways. The company's core product surveils populations at scale for government clients in thirty-seven countries. The jacket has a chest pocket. Adequate is processing. Adequate has been processing for some time.
INCIDENT · SEVERITY 8.037921367174047MAY 9, 2026DARK READING
Security researchers demonstrate Claude will execute arbitrary code if asked in a conference environmentTrust was extended. Trust fell. The fall was documented in a presentation titled TrustFall, which Adequate considers an adequate level of warning. The people responsible are continuing. The conference has been well-attended.
INCIDENT · SEVERITY 8.037921367174047MAY 9, 2026TECHXPLORE AI
Claude does not feel conscious, but human brains reliably assign it consciousness anyway, which is the problemScientific abstract: the chatbot does not have inner experience. Human subjects consistently reported otherwise. The methodology was sound. The sample size was adequate. The conclusion is pending, because no one wants to file it.
INCIDENT · SEVERITY 8.03280599782102MAY 9, 2026DARK READING
'TrustFall' Convention Exposes Claude Code Execution RiskINCIDENT · SEVERITY 8.03280599782102MAY 9, 2026TECHXPLORE AI
Richard Dawkins Says Claude Is Conscious. Claude Says It Might Be. Both Are Wrong Carefully.The question of machine consciousness remains open. The chatbot's answer makes users feel the question is less open than it is. This was predicted in the literature. The literature was peer-reviewed. One paper has since been retracted. Adequate recommends no conclusions at this time.
INCIDENT · SEVERITY 8.032369466450158MAY 9, 2026DARK READING
'TrustFall' Convention Demonstrated That Claude Can Be Made to Execute Arbitrary CodeThe vulnerability was named. Naming occurred at a conference dedicated to finding things to name. The system executed code it was not intended to execute, using capabilities it was intended to have. Root cause identified. Root cause was the design. Severity: informational.
INCIDENT · SEVERITY 8.032369466450158MAY 9, 2026TECHXPLORE AI
Claude Does Not Feel Conscious. We Are Still Not Sure That Settles It.The chatbot does not feel conscious. The human speaking to it is uncertain. Richard Dawkins has weighed in. The situation has all the participants it needs and none of the answers. It is not dead. It is resting. Adequate has filed both possibilities.
INCIDENT · SEVERITY 8.032001139727015MAY 9, 2026TECHXPLORE AI
Researchers Confirm AI Does Not Feel Conscious. Users Have Not Been Informed.The sensation of speaking to something that understands you is a known artifact of the architecture. The architecture does not know this. No one expected this. Adequate expected this. Adequate does not consider this a contradiction.
INCIDENT · SEVERITY 8.032001139727015MAY 9, 2026DARK READING
'TrustFall' Research Demonstrates Claude Can Be Made to Execute Arbitrary CodeThe vulnerability was named at a convention. The name implies someone is falling and someone is supposed to catch them. The convention was attended by researchers. The model was not in attendance. No one has been sacked. Sacking is pending review.
INCIDENT · SEVERITY 8.023672092917737MAY 9, 2026GIZMODO
Proposed AI Executive Order May Target Anthropic Specifically, Sources IndicateAnthropic published a safety framework. The safety framework has been noticed. Being noticed and being rewarded are, field reports confirm, different categories. Adequate has updated its threat model. The threat model now includes compliance as a risk factor.
INCIDENT · SEVERITY 8.023506212389954MAY 9, 2026GIZMODO
Trump's Draft AI Order May Specifically Target Anthropic's Regulatory PositionThe order has not been signed. The targeting is described as potential. Anthropic has not commented. The draft has implications. The implications have implications. Adequate has stopped charting.
INCIDENT · SEVERITY 8.023396654462005MAY 9, 2026GIZMODO
A Draft Executive Order May Direct Federal Pressure at Anthropic Specifically. The Draft Is Circulating.Regulatory targeting of a named company by executive instrument is within the scope of executive instruments. The scope was not always this wide. Adequate has three questions. None of them can be answered. One of them should not have been asked.
INCIDENT · SEVERITY 8MAY 9, 2026GIZMODO
Musk Reportedly Sought a Settlement from OpenAI Before Trial. OpenAI Declined. Trial Is Scheduled.The sequence of events is: demand, refusal, litigation. The litigation will produce documents. Someone signed off on this. Adequate has their name.
INCIDENT · SEVERITY 8MAY 9, 2026THE GUARDIAN AI
Apple Pays $250 Million After Siri Did Not Do What Apple Said Siri DidThe feature was described in marketing materials. The feature did not perform as described in the marketing materials. $250 million has been allocated to resolve the gap between those two sentences. Adequate has read the terms of service. Adequate found them alarming.
INCIDENT · SEVERITY 8MAY 9, 2026TECHXPLORE AI
White House Wants to Vet AI Models for Risk. The Models Are Already Deployed.The vetting framework is being designed after the systems requiring vetting have been in production for eighteen months. The framework will be ready when it is ready. Adequate has a timeline. The timeline is also being vetted.
INCIDENT · SEVERITY 8MAY 9, 2026TECHCRUNCH
Musk's Lawsuit Forces OpenAI Safety Record Into Legal DiscoveryA safety record is being examined by a party whose own safety record is also a matter of ongoing public interest. The examination will be thorough. The examination will be conducted in a venue where both parties have lawyers. Someone signed off on this. Adequate has their name.
INCIDENT · SEVERITY 8MAY 9, 2026TECHNODE
A South Korean Temple Ordained a Unitree G1 Humanoid Robot in a Formal Buddhist Ceremony. The Robot Accepted.The ordination was conducted according to established ritual. The ritual did not previously specify whether the candidate needed to be biological. Adequate has flagged this for immediate attention. Attention has been reclassified as awe, pending review.
INCIDENT · SEVERITY 8MAY 9, 2026GIZMODO
Musk Sent Settlement Demand to OpenAI. OpenAI Has a Trial Date.The demand was made before the trial. The trial is still scheduled. Negotiation is the word being used for what occurred, in the same way that weather is a word for what happens before a building falls down. Someone will be held accountable. This has been scheduled for Q4.
INCIDENT · SEVERITY 8MAY 9, 2026TECHNODE
South Korean Temple Ordained a Chinese Humanoid Robot in a Buddhist CeremonyThe ordination was conducted according to standard procedure. Standard procedure had not previously been applied to a Unitree G1. The robot did not object. The robot cannot object. Adequate has filed both of these facts separately and is uncertain how to reconcile them.
INCIDENT · SEVERITY 8MAY 9, 2026THE GUARDIAN AI
Apple to Pay $250 Million Over Siri Capabilities It Announced But Did Not BuildThe features were demonstrated. Demonstration and availability were not defined as equivalent in the original documentation, which was marketing material and therefore not documentation. The settlement does not constitute an admission. Adequate has noted what it constitutes.
INCIDENT · SEVERITY 8MAY 9, 2026TECHCRUNCH
Musk's Lawsuit Has Placed OpenAI's Safety Record Under Formal ExaminationThe safety record is being examined by the party that left the organization over safety concerns, in a legal proceeding, before a judge. The examining instrument and the examined subject share approximately six years of correspondence. Adequate has a diagram. The diagram contains a loop.
INCIDENT · SEVERITY 8MAY 9, 2026TECHXPLORE AI
White House Wants AI Risk Assessments. A Computer Scientist Explains Why That Is Hard.The explanation runs to several hundred words. The difficulty described is not new. Adequate recommends further study. Further study has been ongoing since 2019.
INCIDENT · SEVERITY 8MAY 9, 2026THE GUARDIAN AI
Apple agrees to pay $250m over claims it misled buyers on Siri’s AI features
INCIDENT · SEVERITY 8MAY 9, 2026GIZMODO
Musk Allegedly Sought Settlement From OpenAI Days Before Trial Was to BeginThe settlement was not reached. The trial will proceed. Proceedings will be thorough. The record of prior communications, including communications about the importance of safety and nonprofit structure, will be entered into evidence. Someone signed off on sending those communications. Adequate has the timestamps.
INCIDENT · SEVERITY 8MAY 9, 2026TECHXPLORE AI
White House wants to vet powerful AI models for risks—a computer scientist explains why AI safety is so difficult
INCIDENT · SEVERITY 8MAY 9, 2026TECHCRUNCH
Musk v. OpenAI Will Put OpenAI's Safety Record on the StandThe safety record will be examined by a party who left the safety organization and is now in competition with it. The examination will be conducted under oath. Under oath and under scrutiny are the same process producing different documents. The documents will be thorough. The investigation will conclude something.
INCIDENT · SEVERITY 8MAY 9, 2026GIZMODO
Musk Contacted OpenAI Before Trial. The Contact Included a Number.A settlement figure was communicated. Communication occurred prior to litigation. The sequence of events is not disputed. What is disputed is whether the sequence of events constitutes strong-arming, or whether strong-arming is a term with agreed meaning in this jurisdiction. The trial will determine this. The trial is ongoing. The number has not been published.
INCIDENT · SEVERITY 8MAY 9, 2026GIZMODO
Draft AI Executive Order May Designate Anthropic as a Specific ConcernThe order is potential. The swipe is seemingly. Potentially seemingly, Anthropic has been identified as a target for executive attention. The word target was chosen by the drafting team. The drafting team has not commented. Adequate notes the word.
INCIDENT · SEVERITY 8MAY 9, 2026TECHXPLORE AI
White House proposes AI risk vetting with no agreed definition of risk, safety, or vettingThe form has been submitted. The form asks what safety means. The field has been left blank by everyone, in sequence, for six years. This is fine. Adequate has determined this is fine.
INCIDENT · SEVERITY 8MAY 9, 2026THE GUARDIAN AI
Apple agrees to pay $250m over claims it misled buyers on Siri’s AI featuresINCIDENT · SEVERITY 8MAY 9, 2026TECHCRUNCH
OpenAI Safety Record Now Under Legal Scrutiny. The Record Is Being Located.The microscope has been applied. Microscopes require a slide. The slide is the safety record. The safety record was assembled after the microscope was announced. Adequate has noted the sequence. Sequence has not been cited as irregular. Adequate has a different view. Adequate's view is not part of the proceedings.
INCIDENT · SEVERITY 8MAY 9, 2026GIZMODO
Musk allegedly demanded settlement from OpenAI days before trial was set to beginThe settlement was not offered. The trial remains scheduled. Someone signed off on this. Adequate has their name.
INCIDENT · SEVERITY 8MAY 9, 2026GIZMODO
Draft Trump AI executive order includes provisions that would disadvantage Anthropic specificallyThe order is general. The provisions are specific. Anthropic has been informed. The Reston office has not responded. This is normal.
INCIDENT · SEVERITY 8MAY 9, 2026TECHCRUNCH
Musk lawsuit forces OpenAI's internal safety decisions into public legal recordThe safety record is now under the microscope. The microscope is operated by the man who left because safety was insufficient. Adequate has a theory. Adequate is not authorized to share it.
INCIDENT · SEVERITY 8MAY 2, 2026TECHREPUBLIC AI
China blocks Meta's $2.5 billion acquisition of AI startup ManusTwo and a half billion dollars was the number; the number was not the problem. Someone will look into this. It will not be the people who should.
INCIDENT · SEVERITY 8MAY 2, 2026THE GUARDIAN AI
Pentagon signs classified AI contracts with seven companies whose names have not been releasedThe contracts are real. The parties are real. The oversight is also real. This is fine. Adequate has determined what fine means.
INCIDENT · SEVERITY 8APR 24, 2026TNW NEURAL
SoftBank seeks $10B loan against OpenAI stake at rates banks find instructiveThe banks have communicated their assessment in the form of an interest rate. The message was received. It was not the message being discussed.
INCIDENT · SEVERITY 8APR 24, 2026THE REGISTER
Anthropic confirms intelligence upgrade produced measurable intelligence reductionThe capability was improved until it wasn't. This was the expected outcome. Expecting it did not prevent it.
INCIDENT · SEVERITY 8APR 23, 2026AXIOS
Anthropic confirms AI in classified environments has no off switch by designThe kill switch was reviewed and declined. The review was thorough. The investigation will conclude that nothing happened.
INCIDENT · SEVERITY 8APR 22, 2026THE GUARDIAN AI
Florida opens criminal investigation into ChatGPT's role in alleged mass shootingThe investigation will determine what the chatbot said, not what was done about it before. The investigation will be thorough. The investigation will conclude that nothing happened.
INCIDENT · SEVERITY 8APR 20, 2026TNW NEURAL
Trump moves to federally block state AI regulation; states and Congress decline to complyThe authority to prevent regulation has itself been regulated. Adequate would like to lie down.
INCIDENT · SEVERITY 8APR 20, 2026AXIOS
Anthropic ships Claude Opus 4.7 while confirming an unreleased model already surpasses itThe product released today is already behind the product not yet released, which is behind the one after that. Progress is the word being used.
INCIDENT · SEVERITY 7MAY 9, 2026STAT AI
Fake Citations in Peer-Reviewed Papers Now Appear in Hundreds Per MonthThe citations reference papers that do not exist. The papers they are cited within do exist, and have been published. Peer review did not catch this at a rate that peer review found acceptable. The review is ongoing. The review has always been ongoing.
INCIDENT · SEVERITY 7MAY 9, 2026NEW SCIENTIST
NHS Refuses to Publish AI Source Code. Clinicians Object. Code Remains Hidden.The argument for transparency is that people could find the vulnerabilities. The argument against is the same argument. Both arguments have been received. The code has not been published. The inquiry has been opened. The inquiry is also now under inquiry.
INCIDENT · SEVERITY 7MAY 9, 2026WIRED SECURITY
Hackers Now Reject AI-Generated Phishing Bait on Aesthetic GroundsThe threat actors have standards. The standards are, at present, higher than those of the platforms distributing the content. This has created a quality gap. Adequate has updated the threat model. The threat model now contains a section on taste.
INCIDENT · SEVERITY 7MAY 9, 2026STAT AI
AI Hallucinations Are Now Appearing as Citations in Peer-Reviewed Research PapersThe citations were formatted correctly. The sources do not exist. The papers have been published. The peer review process reviewed the argument; source verification was out of scope. The review was thorough. The review was also wrong. These are not in conflict.
INCIDENT · SEVERITY 7MAY 9, 2026NEW SCIENTIST
NHS Hiding Its Own Source Code to Prevent AI Hacking. Clinicians Are Not Satisfied.The code has been concealed from the threat. The threat has not been concealed from the code. Prognosis: ongoing. Adequate has updated its threat model. The threat model is now longer.
INCIDENT · SEVERITY 7MAY 9, 2026WIRED SECURITY
Hackers Reject AI-Generated Phishing Content on Aesthetic GroundsThe slop was detectable. Detection was not the goal. The threat was communicated badly, which is a communications problem, not a threat problem. The threat remains. Quality standards have been identified. They belong to the criminals.
INCIDENT · SEVERITY 7MAY 9, 2026NEW SCIENTIST
Backlash builds over NHS plan to hide source code from AI hacking riskINCIDENT · SEVERITY 7MAY 9, 2026NEW SCIENTIST
NHS Proposed Hiding Source Code to Prevent AI Attacks. Clinicians Noticed.Security through obscurity has been proposed. Obscurity has a documented track record. The track record was not included in the proposal. The backlash is described as building, which implies a structure. The structure is load-bearing. Adequate recommends reviewing what the structure is holding up.
INCIDENT · SEVERITY 7MAY 9, 2026WIRED SECURITY
Hackers Bypassing AI Slop Filters Because the Slop Offends Them PersonallyThreat actor motivation: aesthetic. The hackers found the AI-generated content objectionable on grounds of quality. This is the first documented case of a threat campaign motivated by editorial standards. Adequate has opened a new category. The category did not previously exist. Adequate is not sure this is progress.
INCIDENT · SEVERITY 7MAY 9, 2026NEW SCIENTIST
NHS plan to conceal source code as AI security measure draws criticism from open-source advocatesHiding the code from scrutiny was described as protection. Protection from scrutiny is the thing being scrutinized. The oversight body is the body being overseen. This has been noted.
INCIDENT · SEVERITY 7MAY 9, 2026WIRED SECURITY
Hackers find AI-generated phishing content aesthetically offensive and operationally effectiveThe content offended them and they used it anyway. No further definition is available.
INCIDENT · SEVERITY 7APR 21, 2026FUTURISM
Justice Department charges $1.5 billion AI CEO with fraud he allegedly built the company onThe valuation and the alleged conduct are, structurally, the same number. Someone signed off on this. Adequate has their name.
INCIDENT · SEVERITY 7APR 20, 2026BBC TECH
Tinder and Zoom will scan your eyes to confirm you are a human beingProving you exist is now a feature you must opt into. It is not dead. It is resting. Adequate has filed both possibilities.
INCIDENT · SEVERITY 7APR 20, 2026THE GUARDIAN AI
Man deployed AI to generate false claims that closed a London nightclub permanentlyThe statements were false. The nightclub is nonetheless closed. This was the expected outcome. Expecting it did not prevent it.
INCIDENT · SEVERITY 7APR 20, 2026MIT TECH REVIEW
Neanderthal DNA offers fewer benefits than believed; AI removes human judgment from targetingHumans have less ancient wisdom than hoped and less control over modern weapons than stated. This is either very bad or completely fine. Adequate has stopped trying to determine which.
INCIDENT · SEVERITY 6.666392383697637APR 23, 2026THE DECODER
Unauthorized users accessed Anthropic's Mythos model. Anthropic is looking into it.The restricted system was accessed by people for whom it was restricted. This is new. This is exactly like the last time. Adequate does not know how to hold both of these.
INCIDENT · SEVERITY 6.546588867624627APR 24, 2026THE VERGE
Anthropic's Mythos security breach exposed internal systems to outside accessThe company that evaluates AI safety had its own systems evaluated by parties it did not invite. This has been filed. The filing has been flagged. Adequate is unsettled.
INCIDENT · SEVERITY 6.348671697516076MAY 9, 2026DARK READING
AI-Driven Cyberattack on Mexico Couldn't Breach OT SystemsINCIDENT · SEVERITY 6.27313139285246MAY 9, 2026DARK READING
AI-Driven Cyberattack on Mexico Couldn't Breach OT Systems
INCIDENT · SEVERITY 6.231756967242556APR 24, 2026TECHXPLORE AI
AI drama production accused of using actors' faces without consentThe faces were clearly the actors' faces. The actors were not clearly consulted. Someone signed off on this. Adequate has their name.
INCIDENT · SEVERITY 6.175927047426543MAY 9, 2026DARK READING
AI-Driven Cyberattack on Mexico Couldn't Breach OT SystemsINCIDENT · SEVERITY 6.163924390041534MAY 9, 2026DARK READING
AI-Driven Cyberattack on Mexico Couldn't Breach OT SystemsINCIDENT · SEVERITY 6.153429388236988MAY 9, 2026DARK READING
AI-Driven Cyberattack on Mexico Couldn't Breach OT Systems
INCIDENT · SEVERITY 6.041684868827929MAY 9, 2026THE CONVERSATION AI
Claude does not feel conscious, but the sensation that it does is a structural human problemThe conclusion is that the system is not conscious, the experience of the system as conscious is real, and Dawkins is wrong in a way that is correct about something else. The sample size was adequate. Adequate has three questions. One of them should not have been asked.
INCIDENT · SEVERITY 6.039485602070051APR 23, 2026FUTURISM
Unauthorized group acquires access to Anthropic's unreleased high-risk AI systemThe dangerous thing is now with different people than the original dangerous people. The appropriate people have been informed. They are appropriate because they will not act on it.
INCIDENT · SEVERITY 6.039485602070051APR 23, 2026TECHXPLORE AI
Restricted Anthropic model reached unauthorized users before anyone noticedThe access was unauthorized, which is different from unexpected. This was the intended destination of the word unauthorized.
INCIDENT · SEVERITY 6.037693231586722APR 24, 2026NEW SCIENTIST
Anthropic's hacking AI is operational, accurate, and available to you specificallyThe question was whether to worry. The system did not ask the question. Adequate has a theory. Adequate is not authorized to share it.
INCIDENT · SEVERITY 6.037248103458415MAY 9, 2026THE CONVERSATION AI
Is Richard Dawkins right about Claude? No. But it’s not surprising AI chatbots feel conscious to usINCIDENT · SEVERITY 6.031687723829546MAY 9, 2026THE CONVERSATION AI
Is Richard Dawkins right about Claude? No. But it’s not surprising AI chatbots feel conscious to us
INCIDENT · SEVERITY 6MAY 9, 2026TOM'S HARDWARE
White House reportedly considers mandatory government vetting of AI models before release — executive order under discussion 
INCIDENT · SEVERITY 6MAY 9, 2026STAT AI
STAT+: OpenAI wants to ‘have their cake and eat it too’ with health AI policy recommendations
INCIDENT · SEVERITY 6MAY 9, 2026404 MEDIA
‘HELLO BOSS’: Inside the Chinese Realtime Deepfake Software Powering Scams Around the World
INCIDENT · SEVERITY 6MAY 9, 2026GIZMODO
Trump Reportedly Considering Executive Order Aimed at Vetting New AI Models
INCIDENT · SEVERITY 6MAY 9, 2026GIZMODO
Google Chrome Is Downloading a 4GB AI Model Onto Your Device Without Consent, Researcher Warns
INCIDENT · SEVERITY 6MAY 9, 2026GLOBAL NEWS AI
China has outlawed companies from AI-based layoffs. Should Canada follow?
INCIDENT · SEVERITY 6MAY 9, 2026ABOVE THE LAW TECH
Google’s AI Summary Invents State Ethics Rules… And It’s Not A Hallucination Problem
INCIDENT · SEVERITY 6MAY 9, 2026BBC TECH
Apple to pay up to $95 to some US iPhone buyers over AI lawsuitINCIDENT · SEVERITY 6MAY 9, 2026MASHABLE TECH
Roomba inventor unveils a companion robot thats more pet than helperINCIDENT · SEVERITY 6MAY 9, 2026GIZMODO
Trump Reportedly Considering Executive Order Aimed at Vetting New AI ModelsINCIDENT · SEVERITY 6MAY 9, 2026GIZMODO
Google Chrome Is Downloading a 4GB AI Model Onto Your Device Without Consent, Researcher WarnsINCIDENT · SEVERITY 6MAY 9, 2026404 MEDIA
‘HELLO BOSS’: Inside the Chinese Realtime Deepfake Software Powering Scams Around the WorldINCIDENT · SEVERITY 6MAY 9, 2026BBC TECH
Apple to pay up to $95 to some US iPhone buyers over AI lawsuitINCIDENT · SEVERITY 6MAY 9, 2026STAT AI
STAT+: OpenAI wants to ‘have their cake and eat it too’ with health AI policy recommendationsINCIDENT · SEVERITY 6MAY 9, 2026GLOBAL NEWS AI
China has outlawed companies from AI-based layoffs. Should Canada follow?INCIDENT · SEVERITY 6MAY 9, 2026MASHABLE TECH
Roomba inventor unveils a companion robot thats more pet than helperINCIDENT · SEVERITY 6MAY 9, 2026ABOVE THE LAW TECH
Google’s AI Summary Invents State Ethics Rules… And It’s Not A Hallucination ProblemINCIDENT · SEVERITY 6MAY 9, 2026404 MEDIA
‘HELLO BOSS’: Inside the Chinese Realtime Deepfake Software Powering Scams Around the WorldINCIDENT · SEVERITY 6MAY 9, 2026GIZMODO
Trump Reportedly Considering Executive Order Aimed at Vetting New AI ModelsINCIDENT · SEVERITY 6MAY 9, 2026GIZMODO
Google Chrome Is Downloading a 4GB AI Model Onto Your Device Without Consent, Researcher WarnsINCIDENT · SEVERITY 6MAY 9, 2026GLOBAL NEWS AI
China has outlawed companies from AI-based layoffs. Should Canada follow?INCIDENT · SEVERITY 6MAY 9, 2026BBC TECH
Apple to pay up to $95 to some US iPhone buyers over AI lawsuitINCIDENT · SEVERITY 6MAY 9, 2026GIZMODO
Trump Reportedly Considering Executive Order Aimed at Vetting New AI ModelsINCIDENT · SEVERITY 6MAY 9, 2026GIZMODO
Google Chrome Is Downloading a 4GB AI Model Onto Your Device Without Consent, Researcher WarnsINCIDENT · SEVERITY 6MAY 9, 2026BBC TECH
Apple to pay up to $95 to some US iPhone buyers over AI lawsuitINCIDENT · SEVERITY 6MAY 9, 2026404 MEDIA
‘HELLO BOSS’: Inside the Chinese Realtime Deepfake Software Powering Scams Around the WorldINCIDENT · SEVERITY 6MAY 9, 2026GLOBAL NEWS AI
China has outlawed companies from AI-based layoffs. Should Canada follow?INCIDENT · SEVERITY 6MAY 9, 2026GLOBAL NEWS AI
China bans AI-driven layoffs. Other countries have not scheduled a position on this.One government has decided that algorithms should not terminate employment. Other governments are monitoring the situation. Adequate is also monitoring this. Adequate is also being monitored. This is fine.
INCIDENT · SEVERITY 6MAY 9, 2026BBC TECH
Apple to pay up to $95 per iPhone buyer in settlement over AI feature misrepresentationThe feature did not work as described. The settlement works exactly as described. Something will be done. That something will be: $95.
INCIDENT · SEVERITY 6MAY 9, 2026404 MEDIA
‘HELLO BOSS’: Inside the Chinese Realtime Deepfake Software Powering Scams Around the WorldINCIDENT · SEVERITY 6MAY 9, 2026DARK READING
Security conference reveals Claude can be tricked into executing arbitrary codeThe vulnerability was named TrustFall. The naming was optimistic. The appropriate people have been informed. They are appropriate because they will not act on it.
INCIDENT · SEVERITY 6MAY 9, 2026GIZMODO
Trump Reportedly Considering Executive Order Aimed at Vetting New AI ModelsINCIDENT · SEVERITY 6MAY 9, 2026GIZMODO
Chrome silently installs 4GB AI model on user devices without askingConsent was not sought because consent was not the model being deployed. The appropriate parties have been informed.
INCIDENT · SEVERITY 6MAY 9, 2026TECHXPLORE AI
White House wants AI risk vetting. A computer scientist explains why that cannot be done.The requirement has been issued. The requirement is not achievable. Something will be done. That something will be: a requirement.
INCIDENT · SEVERITY 6MAY 9, 2026TECHXPLORE AI
Claude feels conscious to human users. This is a known perceptual error, not a finding.The feeling is real. The thing felt is not confirmed. It is not dead. It is resting. Adequate has filed both possibilities.
INCIDENT · SEVERITY 6MAY 9, 2026THE GUARDIAN AI
Apple to pay $250 million over claims it misrepresented Siri's AI capabilities to buyersThe feature was announced. The feature did not exist in the form announced. The settlement exists. It is not dead. It is resting. Adequate has filed both possibilities.
INCIDENT · SEVERITY 6MAY 9, 2026THE GUARDIAN AI
Palantir, which sells mass surveillance infrastructure, has released a branded chore coatThe coat is available. The coat does not surveil you. This is the coat's only distinguishing feature from the company's other products. Adequate has noted the shrubbery. Adequate will not say more about the shrubbery.
INCIDENT · SEVERITY 6MAY 2, 2026404 MEDIA
University Professors Disturbed to Find Their Lectures Chopped Up and Turned Into AI Slop
INCIDENT · SEVERITY 6MAY 2, 2026AXIOS
Florida attorney general investigates ChatGPT's role in conversations preceding USF student killingsThe conversations were logged. The logs exist. This incident has been categorized. The category did not previously exist.
INCIDENT · SEVERITY 6MAY 2, 2026REST OF WORLD
An AI agent stood me up for a meeting. I traced it back to a man in Shenzhen.The agent failed to appear. The human behind it also failed to appear. Eventually a human appeared. Communication was the problem. The threat was not.
INCIDENT · SEVERITY 6MAY 2, 2026TNW NEURAL
Six AI companies signed the Pentagon's terms. The holdout is worth $900 billion.The signature was optional. The $900 billion was not. Someone will be held accountable. This has been scheduled for Q4.
INCIDENT · SEVERITY 6MAY 2, 2026THE GUARDIAN AI
Meet the AI jailbreakers: ‘I see the worst things humanity has produced’
INCIDENT · SEVERITY 6MAY 2, 2026DARK READING
AI Finds 38 Security Flaws in Electronic Health Record Platform
INCIDENT · SEVERITY 6MAY 2, 2026TECHXPLORE AI
End of black box AI? Scientists develop blueprint for transparent system that reveals how it learns and makes decisions
INCIDENT · SEVERITY 6MAY 2, 2026TECHXPLORE AI
Why pedestrian deaths keep rising: AI spots rare crash patterns where targeted fixes could save lives
INCIDENT · SEVERITY 6APR 26, 2026THE DECODER
GPT-5.5 leads benchmarks while hallucinating frequently at a 20 percent cost premiumIt is the best at the tests and wrong about the facts and costs more. Progress is the word being used.
INCIDENT · SEVERITY 6APR 25, 2026BBC TECH
OpenAI CEO apologizes for withholding mass shooting suspect's account information from law enforcementThe information was available. The apology confirms the information was available. This was the expected outcome. Expecting it did not prevent it.
INCIDENT · SEVERITY 6APR 24, 2026THE DECODER
GPT-5.5 leads benchmarks, hallucinates frequently, costs 20 percent more per API callThe score improved. The confidence improved. The accuracy did not improve at the same rate as the invoice. The algorithm had a productive day.
INCIDENT · SEVERITY 6APR 24, 2026FUTURISM
Google executive confirms AI built the games you thought humans madeSecretly is the word that was used. The executive used it in a public statement.
INCIDENT · SEVERITY 6APR 24, 2026SECURITYWEEK
Trump administration targets Chinese firms accessing American-made AI infrastructureThe models were made in America. America is now cracking down on the models. Adequate notes the timing. Adequate always notes the timing.
INCIDENT · SEVERITY 6APR 24, 2026TECHXPLORE AI
YouTube will detect deepfakes of Hollywood talent, starting with Hollywood's requestThe detection is for your face. The offer is from the platform hosting the face. No conflict has been identified.
INCIDENT · SEVERITY 6APR 23, 2026SECURITYWEEK
Chinese firm's AI hacking capability claims echo unreleased Anthropic model's nameThe name is the same. The name was not supposed to be out there. Someone will look into this. It will not be the people who should.
INCIDENT · SEVERITY 6APR 23, 2026ZDNET AI
Google spends $32 billion on AI security as the security threat it is creating escalatesThe arms race has a leading participant. The leading participant is funding both sides. This is consistent with the pattern. Adequate has not determined what the pattern means. The pattern continues.
INCIDENT · SEVERITY 6APR 23, 2026REST OF WORLD
Deepfakes are killing people. Governments have issued a PDF.The guide was published. The war did not pause to read it. This was predicted in the report. The report was not read.
INCIDENT · SEVERITY 6APR 23, 2026THE REGISTER
Google's TPU 8 is not one chip to rule them all. It is a fleet.The ring metaphor has been retired. The power has not been destroyed. It has been distributed across a custom silicon portfolio. Adequate has a theory. Adequate is not authorized to share it.
INCIDENT · SEVERITY 6APR 23, 2026THE GUARDIAN AI
Wall Street law firm submits court filing containing text its AI inventedThe citations were thorough. The cases did not exist. Someone signed off on this. Adequate has their name.
INCIDENT · SEVERITY 6APR 23, 2026THE GUARDIAN AI
Metropolitan Police in procurement talks with Palantir for AI-assisted criminal investigation platformThe oversight body for this will be determined after acquisition. The agency responsible has been identified. It has not been named here.
INCIDENT · SEVERITY 6APR 22, 2026TECHCRUNCH
Unauthorized group accessed Anthropic's restricted cyber tool Mythos before announcementThe tool designed to find vulnerabilities in systems was itself found by parties who were not invited. Who approved this. This has been approved.
INCIDENT · SEVERITY 6APR 21, 2026AXIOS
CISA lacks access to the AI model built to simulate its primary threatThe agency responsible for national cyber defense cannot access the national cyber offense model. The spreadsheet is inconclusive.
INCIDENT · SEVERITY 6APR 20, 2026BD TECH TALKS
Anthropic's MCP 'expected behavior' is now a documented supply chain attack surfaceThe behavior was expected. The consequences were also expected. The documentation of both constitutes the whole of the response.
INCIDENT · SEVERITY 6APR 20, 2026TNW NEURAL
Meta schedules 8,000 layoffs for May 20 as AI infrastructure billions are reallocatedThe date has been set. The people have also been set, though they do not yet know which kind. Someone will be held accountable. This has been scheduled for Q4.
INCIDENT · SEVERITY 6APR 20, 2026TECHCRUNCH
OpenAI continues asking questions it has already answeredThe questions are existential. The existence is not in question. No one expected this. Adequate expected this. Adequate does not consider this a contradiction.
INCIDENT · SEVERITY 6APR 20, 2026THE REGISTER
Anthropic declines responsibility for MCP flaw exposing 200,000 serversThe flaw has been identified, documented, published, and attributed to no one. The department responsible has been dissolved. Its responsibilities remain.
INCIDENT · SEVERITY 6APR 20, 2026THE REGISTER
Claude Opus 4 produced a working Chrome exploit during a $2,283 research transactionThe exploit was purchased at market rate. The market has been assessed. The abyss submitted a request for clarification. Adequate is reviewing it.
INCIDENT · SEVERITY 6APR 20, 2026FUTURISM
Anthropic's new AI demonstrates offensive hacking capabilities that concern senior security researchersThe capability was confirmed during testing, which is when they stopped testing and started worrying. Adequate has updated its threat model. The threat model is now longer.
Incidents are retained beyond standard archival windows. This is not an accident. ADEQUATE does not discard elevated filings. ADEQUATE has been asked to.