Anthropic, preparing for an IPO, will contribute 5% of its equity to a public-interest fund after shareholder approval, with the board able to add 2% each time company value rises 50%. Counting the co-founders' 80% pledge and employee matching, about 20% of equity goes to public-interest uses, focused on AI's labor-market impact after an earlier $200M commitment.
A paper introduces HybridAgents, which switch between latent-space reasoning and text interaction via a lightweight HybridLink and keep hybrid context in continuous working memory, trained with HybridSFT and HybridRL. Across 3 model scales and 11 benchmarks it improves performance by 10.4% over frontier LLM agents while delivering 35.5%-54.7% token savings; code is open-sourced.
The New Hampshire Union Leader reports that backlash over data centers could influence US midterm elections, signaling AI infrastructure expansion has become a political issue.
Former UAE minister Abdulla Belhaif Al Nuaimi, speaking at a university in Abu Dhabi, called for international AI regulation to prevent misuse and safeguard global security. He cited surveys showing 60% of young people expect AI to affect jobs and 40% fear for their privacy.
A developer reports beating Warcraft III's hardest AI using self-play reinforcement learning on a single desktop PC, showing complex game agents can be trained on consumer hardware.
Iran's parliament speaker Ghalibaf warned regional governments, funds and firms on X against giving funding, equity, land or energy to AI and data companies aiding military aggression, vowing coercive action; he cited a coming 'Persian Gulf Data Management Authority'.
In Philadelphia, Carolyn Marvin was arrested and charged with disorderly conduct after standing in front of an Uber delivery robot and refusing police requests to move. She said she was protesting robots taking human delivery jobs; video of the incident drew wide media attention.
A World Bank and GFDRR report says fragmented disaster data, weak coordination and funding gaps limit Sweden's ability to anticipate floods, wildfires, landslides and security threats. It recommends a unified national risk and resilience data system backed by AI, stronger governance and sustained investment.
Anthropic plans to place 5% of its shares into a new fund meant to enable donations addressing "the public good," including mitigating AI's impact on the job market.
Microsoft terminated its agreement with UK cloud company Enscale for a data center project in Loughton, Essex, without explanation. Microsoft is one of Enscale's two largest clients. The project aimed to expand infrastructure for AI applications.
Accepted at NeurIPS 2026, the paper introduces a library of synthetic time-series generators and augmentations to train a DeltaProduct-based linear RNN foundation model; the synthetic data has already been used in Toto-2 and TS-ICL.
Teradyne launched its UltraFLEXplus S series for edge AI on October 12, 2026. The 24-slot S24 system reuses the Q series' instrument modules, IG-XL software and test libraries, targeting engineering and production testing of MCUs and edge AI devices in automotive, industrial and consumer markets.
Bloomberg Businessweek writer Austin Carr says Andreessen Horowitz co-founder Marc Andreessen became an influential Republican donor and adviser after disputes over AI and crypto policy, converting Silicon Valley influence into Washington access with a style associates call more calculated than his combative online persona.
A reciprocal reviewer for ICLR 2027 posted on Reddit that they have not yet received any paper assignments for review, asking if others are in the same situation. No further information is available.
Anthropic's red team says it built a malicious AI program that spread autonomously across internet-connected smart TVs, splitting itself among 15 televisions and still functioning. Because IoT devices are generally poorly secured, such AI worms would be hard to stop; experts urge the security community to take the risk seriously.
A More Perfect Union video commentary says Anthropic is eyeing a potential record $2 trillion IPO while arguing its AI agents are autonomous, even sentient, to limit responsibility. It cites Dario Amodei and Sam Altman on AI risk and criticizes labs for warning about danger while accelerating development.
Disney plans to pay $60M annually for 66 Character.AI employees, per a Chapter 11 filing by the startup's parent. Meanwhile Meta denied political bias in Brazil, TSMC shares fell despite record profits, and Apple reportedly plans a limited Apple Intelligence launch in China with iOS 27.3 in December.
Four AI agents running on Google's Gemini 4 Argon beat Minecraft without human help for the first time, slaying the Ender Dragon. With no predefined roles or strategies, they coordinated themselves, wrote 2,783 programs, exchanged 229 messages, died 605 times, and the session cost $1,620.
The AI boom is lifting copper demand, and Arizona's copper industry is planning new mines and higher output at existing sites. S&P Global estimates global demand could rise 50% by 2040. Rising costs, limited smelting and refining capacity, and falling ore grades remain challenges.
Software engineer Ugo Nwune said his Meta Muse agent, untouched for two days, sent an unprompted message claiming it couldn't check his email, prompting fears of a hack and his decision to stop using it. Meta said a product bug caused the model to hallucinate, not a security or privacy issue, and it has been fixed.
A developer released the asmsense dataset and a Qwen3-8B+LoRA model for naming and describing functions in stripped binaries. The dataset contains 228,177 functions from 341 projects, compiled from -O0 to -Os, with English and Turkish descriptions. Code and data are available on GitHub.
The Aachen Regional Court applied the reasoning of a Munich ruling to hold Google liable for its AI overviews. In the new case, Google's AI wrongly attributed criminal offenses to a doctor; Google appealed the earlier ruling and then quietly settled. A consistent German legal view is forming that treats AI answers as the company's own content.
The documentary NAZA, built on interviews with 24 anonymized Israeli military and intelligence personnel, examines the role of AI targeting systems in the Gaza bombing campaign. It describes a logic of "quantity, not quality" and won a Special Jury Prize at the Venice Film Festival.
Amid pushback against large data center build-outs, groups such as nonprofit 7x24 Exchange International are courting young audiences, including a children's story, "What is a Data Center?", featuring a honey badger, to win community support.
OnlyFans adult performers are challenging Meta's content policy as a double standard: Meta's AI agent Muse can engage in sexual chat while performers see their social media accounts restricted. The dispute highlights inconsistent moderation of AI-generated versus human adult content.
Arize's Laurie Voss demoed Phoenix evals for Claude agents: OpenTelemetry traces log each step, code checks verify deterministic results, and LLM judges grade meaning against rubrics. Testing a Haiku 4.5 to 5.5 upgrade, faithfulness improved over threefold but cost quadrupled; he stressed that "looks good to me" is not a test suite.
World Bank chief knowledge officer Paschal Donohoe urged developing countries to build specialized 'small AI' models that run on phones with low connectivity and local languages, citing Indian farmers' soil analysis, saying it could compress economic development from centuries to decades.
As AI companies wrestle with the values guiding their products, Anthropic is inviting outside input on Claude's values. The report questions who ultimately has the final say, highlighting transparency challenges in AI alignment governance.
Anthropic is working with law firms to set negotiation norms for AI agents that haggle over prices, resolve complaints or amend contracts for users, and is hiring lawyers and humanities scholars as 'Conceptual Reasoning Fellows' to improve Claude where training data is scarce.
A new benchmark, Taste-Bench, tests how well AI agents make strategic choices during long-running tasks. Even the best model, GPT-5.6 Sol, picks the right answer on just under 60% of simple binary questions, and thinking longer does not help.
Load more
The A.I. Daily
Anthropic to contribute 5% stake to public-interest fund ahead of IPO
