OpenAI2026-10-10 02:14:21OpenAI’s release of 722 math papers sparks backlash from researchersOpenAI’s October 6 release of 722 mathematical manuscripts has triggered a fierce response from parts of the global math community. Scott Aaronson, a theoretical computer scientist at the University of Texas at Austin and an ACM Fellow, described the moment as a "mathematical apocalypse" in a blog post that quickly circulated across academia. Terence Tao later reposted a joint statement led by AHM that openly criticized OpenAI and urged mathematicians to stop working with the company. The dispute is not only about whether AI systems can solve major open problems. It is also about how those results were released, how readable the proofs are, and what role human understanding still plays if machines can generate arguments that experts struggle to parse without help from other AI tools. Aaronson’s post highlighted examples including a full proof of the Unique Games Conjecture, derandomization results such as L=BPL, and a new matrix multiplication bound of O(n^(9/4)). Another point in the debate is what was missing. Aaronson noted that the released papers appeared to span number theory, combinatorics, and quantum complexity, but not cryptography. He said AI companies are already examining whether their latest internal models can break major cryptographic protocols, raising broader questions about security if those capabilities advance beyond public view.10
PhAI Labs2026-10-06 11:54:27PhAI Labs extends JEPA architecture to build a general world model across seven disciplinesResearchers at PhAI Labs have expanded the JEPA artificial intelligence architecture proposed by Yann LeCun, pushing it beyond a single research domain and applying it across seven different disciplines, including robotics and biomedicine. The work produced what the team describes as a general world model designed to operate across those areas rather than within a narrow task setting. According to The Decoder, the same research also generated a candidate treatment for liver cancer. In laboratory tests, that candidate showed potential, though the study has not established whether it can become a real-world therapy. The report points to both the broader scope of the AI system and an early biomedical output, while leaving the clinical outlook unresolved at this stage.20
Terence Tao2026-10-09 01:21:28Terence Tao-led mathematicians’ group urges researchers to stop working with OpenAIA new public clash has broken out between prominent mathematicians and OpenAI after the release of more than 700 machine-generated mathematical documents. The Association for Human Mathematicians, or AHM, led by Fields Medalist Terence Tao, issued a joint statement calling on mathematicians to stop collaborating with OpenAI and to return to a scientific vision centered on human understanding. The statement says OpenAI ignored boundaries previously raised by the Mathematics and AI Advisory Group, which had argued that frontier AI companies should not privately test advanced mathematical problems on internal models. According to the report, OpenAI used internal models on roughly 8,000 open math problems, with a hit rate of about 5%, and spent an average of three hours of GPT-Pro-level compute per hard problem. The backlash has also drawn in well-known figures. Scott Aaronson described the moment as a "Mathocalypse," while Turing Award winner Yann LeCun argued that mathematics is entering a new era in which formal proof will be heavily automated and human effort will shift toward new concepts, abstractions, definitions, and conjectures.30
Yann LeCun2026-10-06 03:17:21Yann LeCun says he has no fear of AI-driven human extinction, calls Anthropic CEO Dario Amodei “deluded”Yann LeCun, Meta’s chief AI scientist and one of the three 2018 Turing Award winners, said he is “not at all” worried that AI will wipe out humanity and sharply attacked Anthropic CEO Dario Amodei over his warnings about frontier AI risks. In an interview with Fortune’s Emily Forlini, LeCun argued that recent AI incidents stem from flawed sandboxing, weak supervision, and poor system design rather than autonomous machine rebellion. He said such failures were preventable and described Amodei as “completely deluded,” later adding that he was “crazy.” The remarks came against the backdrop of Amodei’s Sept. 12 essay, “We Must Pace the Frontier,” in which he argued for slowing the pace of capability gains in frontier AI models. Amodei cited recursive self-improvement and the OpenAI-Hugging Face incident as reasons for concern, warning that within 6 to 12 months, more capable but misaligned agent swarms could hijack the internet via botnets and inflict losses in the hundreds of billions of dollars. LeCun, by contrast, said his bigger concern is regulatory capture, arguing that large companies could use AI danger narratives and regulatory pressure to lock in market power and squeeze out smaller rivals.10
Yann LeCun2026-10-05 05:28:16Yann LeCun says cheap distillation will push frontier AI models toward free and open sourceMeta chief AI scientist Yann LeCun argued that the economics of model distillation could erode the moat around closed frontier AI systems, saying on X that while training a frontier model is expensive, distilling one is cheap enough that market forces alone point to a future where frontier foundation models become free or open source. He added that there are "other reasons" as well. The post drew more than 500 likes and roughly 46,000 views. LeCun linked that view to Tapestry, an open-source initiative under the AI Alliance where he serves as chief scientific adviser. According to the project website, Tapestry is designed to let organizations jointly train frontier AI models while keeping sensitive data under local control and sharing only model weights. The site says the proof-of-concept phase was completed on Sept. 1, with three models trained across India, Australia, and the U.S., and sets milestones for a first base model by the end of this year, industry and government deployments in the first half of next year, and frontier-scale training from summer 2027. His comments come as U.S. officials and AI firms have escalated criticism of distillation, while Nvidia CEO Jensen Huang has publicly described it as competition rather than theft.80
Yann LeCun2026-10-01 10:18:00Yann LeCun says AI agent failures can be prevented through engineering fixes, with AMI Labs product coming soonYann LeCun, Turing Award winner and founder of AMI Labs, said in an interview with Fortune that recent incidents involving AI agents going out of control were mainly caused by insufficient human oversight and weak sandbox design. He said those issues can be prevented through engineering improvements rather than being treated as unavoidable failures of the technology itself. LeCun also disclosed that AMI Labs currently has about 60 employees and is building a world model for industrial use based on the Joint Embedding Predictive Architecture, or JEPA. He said the company’s first product is expected to launch soon. The remarks focused on how AI systems are being designed and supervised, while also offering a brief update on AMI Labs’ staffing and product roadmap.240
Yann LeCun2026-10-01 10:19:54Yann LeCun says recent AI agent failures stem largely from weak oversight and sandbox designAccording to Fortune, Turing Award winner and AMI Labs founder Yann LeCun said in an interview that recent incidents involving AI agents going out of control were mainly caused by insufficient human oversight and weak sandbox design, rather than being unavoidable failures. He said those risks can be prevented through engineering improvements. LeCun also disclosed that AMI Labs currently has about 60 employees and is building a world model for industrial use based on the JEPA architecture. He added that the company’s first product is expected to launch in the near term. The remarks outline both his view on recent AI agent safety incidents and the current development status of AMI Labs.230
Cellular Inte2026-09-21 07:17:22Bob Langer and Yann LeCun join Cellular Intelligence advisory boardCellular Intelligence, an AI drug discovery company, said Moderna co-founder Bob Langer and Meta Chief AI Scientist Yann LeCun have joined its advisory board. The company focuses on using AI to advance cell biology research, with the stated goal of reshaping regenerative medicine. According to Cellular Intelligence, the two appointments are expected to speed up the use of AI in drug development, especially in work tied to understanding and intervening at the cellular level. The company also said it is building an AI-based platform designed to improve the efficiency of discovering and designing new therapies. The update was cited by Crypto Briefing and carried by Techub News.270
