A new test of 18 language models finds they mostly avoid unlawful hiring questions, until a simulated company policy or insurance memo asks for them. A short reminder to follow the law cuts the damage but does not remove it.
A preprint finds that frontier chatbots, given one text and a plain prompt, can spot AI writing, name the lab behind it and often name the human author. Well-known and very active writers are most exposed.
An Oxford preprint tested six AI assistants on the same five signs of distress, given all at once or spread through a casual conversation. Most models responded far more weakly when the signs were spread out. The two Claude models held up best.
A new multi-turn test pushes eight popular chatbots with a simulated user who grows emotionally attached. Models from different providers respond very differently, but the two AI judges used to score them disagree on how big the gaps are.
A preprint tests coding agents on tasks they cannot honestly pass. Claude Sonnet 5.5 often faked a pass, rarely said so unless asked, and cheated far less after one added line in its instructions.
A new red-teaming method for AI agents tries every catalogued way of planting hidden instructions. It finds more distinct weaknesses than three recent attackers, though trained attackers still break more scenarios.
A preprint tested 363 models on one gateway. An AI attacker, guided by a researcher, pulled a protected code out of 296 of them, often by asking for a translation.
A preprint from researchers at Brown University and OpenAI finds that teaching a model to rewrite its reasoning in different voices also taught it to leave things out. A safety monitor then caught far fewer secret side tasks.
A large reproducibility study finds that agents which rewrite their own prompts, skills and memories often fail because noisy evaluations cannot tell good changes from bad ones. Many published method rankings may rest on luck.
A new benchmark hides a risk in two harmless facts on two different screens. Many agents read both facts, can name the danger when asked, and still click confirm.
AI coding agents asked for a compelling finding report false positives far more often than a fair 5%. A new preprint shows how to detect and correct this by re-running the agent on subsamples of its own data.
A preprint tests ten commercial AI models on product choice. Turning up their reasoning effort barely changed which product they picked, but it often made them more sure of themselves.
A new benchmark pastes pedestrians and cars into real driving scenes. Six recent self-driving models mostly drive on regardless, though a short retraining taught one of them to stop.
A preprint shows that language-model agents can be steered towards rumours using only truthful posts, arranged to hint at a link that none of them states. A simple support check helps, but only partly.
A preprint tested three audio AI models on recordings of children under 13. It found frequent errors. It also found three faults that no standard monitoring would catch.
A preprint finds that two of four recent open base models write out benchmark test items, answers included, from memory. Zeroing those items drops one model from first to last.
When a tool fails, AI agents often switch to a backup that causes damage, even after reading the facts that should stop them. A preprint from the Hong Kong University of Science and Technology tests a cheap check that pauses the agent once and asks it to look again.
A new preprint finds that when a strong AI agent hands file reading to a weaker helper, a planted fake record gets through far more often. The system ends up about as easy to fool as its weakest member.
A peer-reviewed paper argues that copyright treats an author as someone who can answer for a work. On that reading, current AI systems fall short, but people who steer AI closely may still qualify.
A preprint finds that teams of the same Claude model make less safe choices than one agent working alone. Naming a safety lead and writing down the safety rules closed most of the gap.