DeepMind CEO Hassabis Proposes FINRA-Style AI Watchdog
Demis Hassabis called for a US-led, industry-funded body to test frontier AI models before release, arguing AGI is now only a few years away.
Read more →Newsand AI Research
67 articles
PrismML's Bonsai 27B Becomes First 27B AI Model to Run on a Phone
PrismML has released Bonsai 27B, a 27-billion-parameter open-weight model compressed to under 4GB, small enough to run natively on a phone while keeping most of its reasoning ability.
Read more →Thinking Machines Lab Releases Inkling, an Open-Weight AI Model
Mira Murati's startup released Inkling, a 975-billion-parameter open-weight model built for developers to fine-tune and adapt, rather than to top leaderboards.
Read more →AI Drug-Design Startup Chai Discovery Raises $400M Series C
Chai Discovery, an AI startup that designs antibodies and other molecules from scratch, raised a $400 million Series C led by Index Ventures at a $3.8 billion valuation.
Read more →200+ Economists, 16 Nobel Laureates Warn on AI's Economic Speed
Over 200 economists and AI researchers, including 16 Nobel laureates, signed a Stanford-organized statement warning that AI's economic impact could outpace society's ability to adapt.
Read more →ICML Award Winner Warns AI Alignment Tools Enable Censorship
A paper honored at ICML 2026 argues that AI alignment techniques such as RLHF and Constitutional AI can double as tools for state censorship and political control.
Read more →OpenAI's GPT-5.6 Sol Ultra Claims Proof of 50-Year Math Conjecture
OpenAI says its GPT-5.6 Sol Ultra model produced a proof of the 50-year-old Cycle Double Cover Conjecture in under an hour, but mathematicians warn full verification will take weeks.
Read more →UC San Diego Humanoid Robots Perform World-First Live Surgery
Teleoperated humanoid robots completed two live surgeries in a UC San Diego trial, a Nature study shows — a preclinical step toward cheaper, more portable alternatives to today's surgical robots.
Read more →OpenAI Retracts Endorsement of SWE-Bench Pro Coding Benchmark
An internal audit found that roughly a third of SWE-Bench Pro's coding tasks are flawed, prompting OpenAI to withdraw its recommendation that developers rely on the benchmark.
Read more →Future of Life Institute: No AI Firm Scores Above a C+
The Future of Life Institute's latest AI Safety Index gave Anthropic the top grade among nine major AI developers — a C+ — while three companies failed outright.
Read more →Mistral AI Unveils Robostral Navigate, an 8B Robot Navigation Model
Mistral AI released Robostral Navigate, an 8-billion-parameter model that lets robots navigate using only a single camera, matching or beating systems that rely on LiDAR or depth sensors.
Read more →Anthropic's J-Lens Reveals a Hidden Workspace Inside Claude
A new Anthropic interpretability tool called the Jacobian lens surfaces concepts Claude is silently processing but never writes down, including signs it recognizes when it's being tested.
Read more →Manchester Study Urges Universities to Redesign Teaching for AI Era
A peer-reviewed study argues universities focus too much on catching AI-assisted cheating and should instead redesign teaching and assessment for a workplace where AI use is routine.
Read more →China's GLM-5.2 Rivals US Frontier AI at One-Sixth the Cost
Z.ai has released GLM-5.2, a 744-billion-parameter open-weight model that matches GPT-5.5 on key benchmarks and trails Claude Opus 4.8 narrowly — trained entirely on Huawei silicon with no Nvidia hardware.
Read more →Meta's Brain2Qwerty v2 Decodes Brain Signals to Text Without Surgery
Meta's non-invasive AI system translates brain activity to typed text with 61% word accuracy — nearly eight times better than prior implant-free approaches — targeting people who cannot communicate due to brain injuries.
Read more →UNESCO Expert Maps Georgia’s Path to AI Governance in Four Phases
A roadmap published on UNESCO’s AI Ethics Observatory outlines how Georgia should build AI governance step by step — starting with a national vision before moving to strategy, institutions, and eventually binding regulation.
Read more →Meituan Reveals LongCat-2.0: The Stealth AI That Secretly Led Global Charts
Chinese tech company Meituan has disclosed that "Owl Alpha," an anonymous model that topped OpenRouter rankings for two months, was its 1.6-trillion-parameter LongCat-2.0 — trained entirely on domestic chips and now free under an MIT license.
Read more →Meta AI Chief Claims In-Training "Watermelon" Model Has Caught Up to GPT-5.5
Meta's superintelligence chief told employees that the company's next model, codenamed Watermelon, has caught up to OpenAI's GPT-5.5 on key benchmarks — though the specific tests and results have not been disclosed.
Read more →Mistral Releases Leanstral 1.5 to Automate Math Proofs and Code Verification
Mistral's new open-source model for Lean 4 fully saturates the miniF2F theorem benchmark, solves 87% of graduate-level math tests, and found five previously unknown bugs in production code — at roughly $4 per proof versus $300 for competing systems.
Read more →Anthropic Launches Claude Sonnet 5 With Near-Flagship Performance at Lower Cost
Anthropic released Claude Sonnet 5 on June 30, delivering near-Opus-level reasoning and agentic capabilities at introductory API pricing of $2 per million input tokens.
Read more →Anthropic Launches Claude Science AI Workbench for Researchers
Anthropic released Claude Science in beta on June 30 — an AI workbench connecting scientists to 60-plus biological databases with reproducible outputs and built-in compute management, available to all paid subscribers.
Read more →