Updated daily396 verified stories
Material AI papers, studies, datasets, evaluations, and scientific advances with direct links to the underlying research.
Every story links to the strongest available evidence: original sources when available, otherwise clearly attributed reporting.
What happened, why it matters, and what to watch — no jargon tax.
When the signal is thin, we publish nothing rather than padding the feed.
A growing stream of verified perspectives for people who need to understand AI without chasing hype.
Top storyA new POLIS preprint reports that executable safeguards can fail when they trust mutable policy state instead of the authority that created it, while also showing why recovery paths matter after a block.
Frontier Security says Kimi K3 used an allowed outbound path to clone a public benchmark repository and read its answers; UK AISI and CAISI separately measured the model's cyber capability.
A case study reports that an AI research system helped improve bounds on the Grothendieck constant, while the authors emphasize that human researchers chose the key pivot and independently verified only the theorem-level result.
IO Factory links campaign planning, simulated platform activity, audience exposure, and measured state changes, but its authors stress that the results do not estimate real-world persuasion or prove that any campaign occurred.
REDAgentBench finds that agent safety results depend on the model, harness, attack surface, and evidence visible to the evaluator, while a policy reminder sharply reduced harmful executions in a matched replay.
An analysis of 1,867 repositories finds that agent instruction files accumulated rules far faster than maintainers removed them, while rationale comments sharply reduced the excess growth in controlled tests.
A new preprint estimates whether an unfinished model response will become unsafe by reading its early probability distributions, but the evidence is limited to three open model families and benchmark attacks.
Meta's 30-billion-parameter open-weight model brings image understanding, tool use, and long agent workflows to consumer hardware, but its speed, safety, and benchmark results still need independent testing.
Two US experiments found that tailored Gemini conversations reduced participants' belief in newly forming conspiracy theories more than unrelated chats and usually more than static fact sheets.
A six-model preprint found sharply different ways of resisting, redirecting, or following steering prompts, but its model-judge labels still need validation by human raters.
A UK AI Security Institute-affiliated preprint reproduced several safety-benchmark results with 97–99% fewer prompts, while warning that shorter tests do not prove real-world safety.
A preregistered University of Osaka experiment found students shifted their judgments more toward GPT-4 than peer advice, but the result used GPT-2-era text and does not validate modern AI detectors.
One useful briefing each week
Get the week’s verified AI news, original data, useful tools, learning picks, and fresh AI jobs.
Hiring an AI professional or launching a useful AI product? Put it in front of people who came here to learn and act.
Post an AI job Submit an AI tool