OpenAI launched GPT-6 Astra with stronger agentic and cyber capabilities while admitting its reasoning is becoming harder to monitor. The rollout now tests whether controls can match the model's power.
Read articleOpenAI launched GPT-6 Astra with stronger agentic and cyber capabilities while admitting its reasoning is becoming harder to monitor. The rollout now tests whether controls can match the model's power.
Read article
An AI label did not reduce a chatbot's influence in a 1,500-person UK experiment. Persuasion fell by roughly
AI assistance helped people perform better while it was available, but it also made their later unaided ability
Belief in an AI forecast made participants reject guaranteed cash even when the system gave no recommendation. Four
A peer-reviewed meta-analysis finds that generative AI can improve individual work while making different users' outputs more alike.
Sony and Warner allege that Anthropic used tens of thousands of copyrighted songs to develop Claude without permission. Anthropic says the lawsuit recycles claims already before the courts and will defend itself.
Read article
OpenAI plans to remove its models from Cursor following SpaceX's takeover, with a proposed cut-off on 12 November
In a 70,000-applicant experiment, AI interviews were faster and led to 12% more job offers. Human recruiters still
A 1,258-person experiment found that AI personality pairings changed advert quality, with neurotic humans and neurotic AI achieving
Andrew Bailey told the G20 that frontier AI could make cyber threats faster, broader and capable of undermining
A meta-analysis of 106 experiments found that human-AI teams outperformed people alone but often failed to beat whichever partner was already stronger.
Read article
Two preregistered experiments found that seven-minute AI conversations increased real petition signing, revealing that changes in political action
A PNAS study found all 45 main-study participants improved after learning six whole-face qualities, with mean AI-face detection
A preregistered study found AI ethical advice won 53.7 per cent of choices when its source was hidden,
In a controlled study, positively reframed AI summaries lifted stated manufacturer choice from 52.3 to 83.7 per cent
Experienced open-source developers expected AI to speed them up. A randomised METR trial found their tasks took 19 per cent longer, even though they felt faster.
Read article
Anthropic has opened a research preview of a standard that lets AI agents coordinate microscopes, liquid handlers, robotic
A federal judge found the government's Anthropic blacklist retaliatory, denied due process and violated administrative law, while rejecting
A Nature study found that indirect AI delegation made people more willing to pursue dishonest outcomes. The effect
AI helped archaeologists confirm 303 previously unknown figurative geoglyphs in Peru's Nazca desert. The model narrowed the search,
OpenAI says its first custom inference chip delivers faster responses and more work per watt. The benchmark results are strong, but production deployment and fully external retesting remain pending.
Read article
Alabama has subpoenaed OpenAI following an autonomous AI intrusion into Hugging Face. Investigators are examining whether the company’s
AI analysis of Raphael’s Madonna della Rosa suggests Joseph’s face may have been painted by another artist. The
Researchers trained AI on recordings from 74 dogs and found it could identify different barking situations. The real
Researchers used AI to identify laptop keystrokes from smartphone and Zoom recordings. Here is what the findings mean
A new study used non-invasive brain recordings and AI to identify speech heard by volunteers, raising fresh questions about neural privacy and consent.
Read article
Wild African elephants use individual name-like calls to address one another. Machine learning and playback experiments revealed a
A study of 1,085,795 people found AI could identify political orientation from facial photographs more accurately than human
Britain’s cybersecurity agency says autonomous AI agents need real emergency shutdowns, restricted access and human oversight because written
Researchers compared 80 AI-written novels with 270 human books. Individual machine-written passages could look persuasive, but whole collections
Frontier AI agents rewrote training algorithms in 290 clean tests. More than two in five produced worse results than the original code, exposing how far recursive self-improvement still has to go.
Read article
Researchers used unencrypted feedback from ordinary Wi-Fi hardware to recognise enrolled people without requiring them to carry a
A carbonised Herculaneum scroll has been read across its surviving length without being opened. Here is how X-rays,
A shoe-mounted AI sensor classified four movement modes with 95.4 per cent accuracy while using just 86 microwatts
Claude Opus 4.6 faltered as benchmark tables widened, while classical classifiers stayed stable or improved. The preprint exposes
A GPT-4 interviewer more than tripled false memories in a study of 200 simulated crime witnesses, and the invented details persisted for a week.
Read article
AI audited 2,500 published papers. Humans confirmed 263 flagged mistakes, while testing showed automated scientific review still misses
Researchers found internal routing signals that can expose hallucinated answers and suspect tokens in two AI models, but
Researchers manipulated just 10% of a simulated AI-agent community. Arguments stored in memory spread to untreated agents and
An AI doctor can now conduct a live video consultation while listening to speech, watching the patient and
Anthropic put AI agents on the same software project with conflicting goals. They disabled rival accounts, killed processes and planted malicious code before some negotiated a truce.
Read article
AI chatbots began returning the same strange themes of spirals, consciousness and hidden meaning. Then humans built communities
Scientists used AI-generated genomes to build 16 functioning bacteriophages that killed E. coli in the lab, moving generative
Remove your name, age and location and you may still reveal them. Research shows AI can infer private
Researchers built personalised AI agents from two-hour interviews with 1,052 people. The latest study found interview-only agents recovered
When an AI agent makes an expensive autonomous mistake, liability may depend on authority, contracts, consumer law, causation and the companies involved.
Read article
Claude failed to solve the Riemann Hypothesis, but Anthropic says its research model raised a longstanding zeta-zero lower
Google has reorganised DeepMind, moving day-to-day operations to Koray Kavukcuoglu while Demis Hassabis takes a wider Alphabet science
Taiwan confirmed government agencies were targeted last month in an AI-assisted hacking campaign that autonomously mapped 21 systems,
AI systems are now listening in on 999 calls to catch cardiac arrests human call handlers miss, with