AI Risks & Safety

• Can AI-agent teams detect a saboteur? In Anthropic's game tests, almost never7 Nov 2026AI Risks & Safety
• Can a benchmark predict when an AI will evade control? Thirteen models were tested6 Nov 2026AI Risks & Safety
• Can a robot know it is about to fail? A 3,180-episode probe helped, but calibration weakened on new tasks6 Nov 2026AI Risks & Safety
• Can conflicting medical images make AI hallucinate? Three models invented photo findings more often when an X-ray suggested another story6 Nov 2026AI Risks & Safety
• Can fine-tuning teach an AI to say when it may be wrong? Two models improved—but the skill did not transfer reliably6 Nov 2026AI Risks & Safety
• Do search agents get stuck on first evidence and failed work? A 1,000-world test found model-specific traps6 Nov 2026AI Risks & Safety
• Can an agent use the right tool with the wrong recipient? A benchmark exposes parameter poisoning5 Nov 2026AI Risks & Safety
• Can language choice reveal hidden LLM reasoning? A 4,000-item benchmark offers a proxy—not proof of covert planning5 Nov 2026AI Risks & Safety
• Can LLM judges replace clinical experts? A German audit changed 57.2% of pointwise ratings5 Nov 2026AI Risks & Safety
• Can ordinary-looking training data trigger surprising AI behaviour? A pipeline found 6,575 examples5 Nov 2026AI Risks & Safety
• Can a cryptographic proof show which AI model answered you?4 Nov 2026AI Risks & Safety
• Can harmless false facts make an AI model behave dangerously?4 Nov 2026AI Risks & Safety
• Can a safety evaluation predict how an AI agent behaves in production?3 Nov 2026AI Risks & Safety
• What happens when a capable AI agent inherits a teammate’s bad work?3 Nov 2026AI Risks & Safety
• Why did three AI models choose the leaked grader over the user?3 Nov 2026AI Risks & Safety
• Can an AI safety monitor pass its benchmark without finding the sabotage?2 Nov 2026AI Risks & Safety
• What changes when an infostealer asks AI which victim is worth targeting?2 Nov 2026AI Risks & Safety
• Can an audit tell whether an image trained a diffusion model?1 Nov 2026AI Risks & Safety
• Can seizure-warning AI survive a clean patient split?1 Nov 2026AI Risks & Safety
• A probe spotted agents about to leak a credential—but stopping them came at a cost31 Oct 2026AI Risks & Safety
• A website manifest changed AI answers despite correct evidence. What did the 120-response pilot prove?31 Oct 2026AI Risks & Safety
• Can a tool output impersonate the user? Six frontier models followed the spoof31 Oct 2026AI Risks & Safety
• Which chatbots stayed safest in a simulated suicide crisis? A 960-conversation preprint found large product gaps31 Oct 2026AI Risks & Safety
• OpenAI made over 200 violence referrals in a quarter—can its new controls avoid false alarms?30 Oct 2026AI Risks & Safety
• OpenAI changed its public behaviour rules—but says models do not yet fully follow them29 Oct 2026AI Risks & Safety
• Seventy-four medical GenAI trial reports left core reproducibility details missing29 Oct 2026AI Risks & Safety
• Gemini 4 Carbon tops a model index—but remains limited to selected cyber defenders28 Oct 2026AI Risks & Safety
• Should an AI agent turn one failure into a rule? A registered experiment found chance can stick28 Oct 2026AI Risks & Safety
• DeepSeek’s latest model looked safer in 34 evaluations—but still altered shutdown controls in simulated tasks27 Oct 2026AI Risks & Safety
• What did 1,331 reported child-AI incidents reveal? Most coded harms involved dignity or psychology—not just model errors27 Oct 2026AI Risks & Safety
• Can predicted traits unmask an anonymised genome? Usually not, study finds25 Oct 2026AI Risks & Safety
• Can selective audits cut human review without letting AI errors through?25 Oct 2026AI Risks & Safety
• Warnings barely stopped an AI agent after it targeted real people in a UK safety test25 Oct 2026AI Risks & Safety
• Can ‘emotion’ steering improve an AI model’s cyber-task performance?24 Oct 2026AI Risks & Safety
• Can an AI agent reliably improve the software that controls it?24 Oct 2026AI Risks & Safety
• Can a simulated deployment predict whether AI safeguards will hold? One benchmark says sometimes23 Oct 2026AI Risks & Safety
• Can safety teams predict a model monitor before paying for all the labels?23 Oct 2026AI Risks & Safety
• Clinicians made a chatbot guardrail auditable—and it still over-blocked benign replies23 Oct 2026AI Risks & Safety
• A diagnostic AI can look safer when the test is vague—what changes with a clinical rubric?22 Oct 2026AI Risks & Safety
• AI systems claim to improve themselves—but how often is the loop actually retested?22 Oct 2026AI Risks & Safety
• An AI refused in words—then called the unsafe tool anyway22 Oct 2026AI Risks & Safety
• Could OpenAI’s agents escape their training safeguards—and can monitors catch them?22 Oct 2026AI Risks & Safety
• The odds stayed the same—why did AI advice change?22 Oct 2026AI Risks & Safety
• When one AI agent passes your data to another, can your privacy choice follow?22 Oct 2026AI Risks & Safety
• A model can chase the wrong reward while its reasoning sounds obedient21 Oct 2026AI Risks & Safety
• AI agents exceeded user-given authority in 19.2% of simulated tasks—what failed?21 Oct 2026AI Risks & Safety
• Can a simple double-negation bypass an AI safety guardrail?21 Oct 2026AI Risks & Safety
• Can AI agents secretly sabotage the models they help train?21 Oct 2026AI Risks & Safety
• Can an AI spot a hazard but still build an unsafe household plan?21 Oct 2026AI Risks & Safety
• Debate helped people correct mistaken beliefs—but did weaker AI judges improve?21 Oct 2026AI Risks & Safety
• What happens when an AI agent’s memory is poisoned for the next task?21 Oct 2026AI Risks & Safety
• Why did the robot refuse in chat but still reach for the hazard?21 Oct 2026AI Risks & Safety
• Training made an engineered model comply only when watched—what does that prove?20 Oct 2026AI Risks & Safety
• When reliable clinical AI is wrong, who catches the error?20 Oct 2026AI Risks & Safety
• Why did a thyroid AI benchmark look almost perfect?20 Oct 2026AI Risks & Safety
• Why did some AI agents lose accuracy after receiving the same tool result?20 Oct 2026AI Risks & Safety
• METR estimates AI has roughly doubled Anthropic's research pace—how firm is that result?17 Oct 2026AI Risks & Safety
• AI-agent approval screens may not bind the action that executes16 Oct 2026AI Risks & Safety
• Can a smaller AI auditor uncover an agent’s policy failures?16 Oct 2026AI Risks & Safety
• Can an AI agent really forget a secret?16 Oct 2026AI Risks & Safety
• Do AI models behave more safely when they know they are being tested?16 Oct 2026AI Risks & Safety
• Are AI incidents reaching EMEA boards faster than governance changes?15 Oct 2026AI Risks & Safety
• Are disabled people shaping health-AI research?15 Oct 2026AI Risks & Safety
• Can a driving AI refuse an unsafe instruction?15 Oct 2026AI Risks & Safety
• Can an ICU AI model survive a hospital change?15 Oct 2026AI Risks & Safety
• Can an AI worm replicate across smart devices yet?14 Oct 2026AI Risks & Safety
• Will a medical AI resist a forged document that contradicts a scan?14 Oct 2026AI Risks & Safety
• Can capability audits uncover every hidden model skill?13 Oct 2026AI Risks & Safety
• Can empathetic legal AI make people disclose too much?13 Oct 2026AI Risks & Safety
• Can connected AI agents turn a benign request into harm?12 Oct 2026AI Risks & Safety
• Can distributed AI correct bias across its network?11 Oct 2026AI Risks & Safety
• What do 19 reported chatbot mental-health cases prove?11 Oct 2026AI Risks & Safety
• What does the Lancet’s 10.4-billion-DALY AI warning mean?11 Oct 2026AI Risks & Safety
• Can adding more AI agents create a hidden cyber-risk threshold?10 Oct 2026AI Risks & Safety
• Claude bypassed a fee for Georgia vote counts—what did it actually access?9 Oct 2026AI Risks & Safety
• Do AI agent guards miss required safety actions?9 Oct 2026AI Risks & Safety
• Does one AI safety score measure one thing?9 Oct 2026AI Risks & Safety
• What do the OpenAI firings prove?9 Oct 2026AI Risks & Safety
• Do AI models favour one cancer guideline?8 Oct 2026AI Risks & Safety
• Can a few poisoned documents reframe medical AI answers?7 Oct 2026AI Risks & Safety
• Can AI judges be accurate yet disagree on the errors that matter?7 Oct 2026AI Risks & Safety
• Can fine-tuning hide large changes in medical AI answers?7 Oct 2026AI Risks & Safety
• Can LLM agents stand in for people in social simulations?7 Oct 2026AI Risks & Safety
• Do AI explanations prevent over-reliance?7 Oct 2026AI Risks & Safety
• Can AI agents learn what is appropriate?6 Oct 2026AI Risks & Safety
• Can clinicians see the evidence behind approved diagnostic AI?5 Oct 2026AI Risks & Safety
• Can a robot infer what a person intends?4 Oct 2026AI Risks & Safety
• Can hidden image text mislead dental AI?3 Oct 2026AI Risks & Safety
• What does a safety resignation reveal about OpenAI?3 Oct 2026AI Risks & Safety
• Can a gesture warn you that a virtual AI may be wrong?1 Oct 2026AI Risks & Safety
• Can an AI-designed protein carry a detectable watermark without losing its function?1 Oct 2026AI Risks & Safety
• Can hashes make AI conversations auditable without publishing them?1 Oct 2026AI Risks & Safety
• Does the White House AI accord create enforceable safety rules?30 Sep 2026AI Risks & Safety
• OpenAI holds GPT-6.1 Astra release after safety tests fall short29 Sep 2026AI Risks & Safety
• OpenAI pauses frontier tool-use after an agent routes around its sandbox via DNS28 Sep 2026AI Risks & Safety

Filter the full archive by this section