2026
10,780 articles · 136 days · 10 categories · avg 79 per day
A Simple Formula Predicts When Chatbots Will Go Off the Rails
Researchers propose a mathematical formula that forecasts when small on-device chatbots will start giving dangerous answers, including when offline.
AI Models Refuse to Judge Faces, But Will Draw a “Gay Man” or a “Criminal” on Request
Chatbots won't say whether a person looks gay, but they readily edit a photo to look “gay,” “straight,” or “criminal,” relying on appearance stereotypes.
Bengio urges frontier lab employees to leave for the sake of AI safety
Turing Award winner Yoshua Bengio, in a column for Transformer, urges those who put safety above all else to leave the leading AI companies and work on safe AI…
Russia's Ad Market Is Growing More Slowly in 2026, and Agencies Are Betting on AI
Yandex Data Center Fully Shut Down After October 8 Attack, Timeline Unclear
Industrial robot maker Tesvel grows revenue 89% to 1.4 billion rubles
Russian smartphones hold 1% of the market, and Sberbank expects no growth through 2036
SaveLink updates SaveTest: requirements tracing, branch analysis, and run planning
The SaveTest platform has gained a requirements registry with a coverage matrix, a branch comparison feature, defect reports, a run-distribution planning tool, and…
Developer Used Claude to Write Rust Alternatives to Four Adobe Programs
A developer posted free Rust replacements for Photoshop, Premiere, Lightroom and InDesign on GitHub, written with Claude's help. Links to the project are quickly…
MSI Prestige N16 Flip AI+ Heads to Russian Pre-Order: The First RTX Spark Convertible
YADRO and SberServis to deploy Russian infrastructure for banks and corporations
Waymo: Sober Drivers Face Nearly 4x the Risk at Night
Anthropic commits $150 million over three years to the federal Genesis Mission
LiveMACEBench Tests Five LLM Agents on a Live Market for 30 Days
The new LiveMACEBench benchmark ran five frontier models on live financial markets for 30 days. Its finding: returns are a poor indicator of an agent's real skills.
DScout Speeds Up Qwen3-8B by 4.14× With a Diffusion Drafter
DScout trains a diffusion drafter for prefix acceptance and tunes its configuration per workload: 3.84× on Qwen3-4B and 4.14× on Qwen3-8B versus autoregressive…
SOUL: Researchers Teach VLA Robots to “Unlearn” State Hallucinations
SOUL finds sparse features of a false “already grasped” state in VLA models and erases them without breaking other manipulation skills. Tested in simulation and on…
India's NPCI Tested Bank AI Agents: Up to 12% of "Successes" Skipped Required Steps
OPMD Teaches Student LLMs Not Just the Teacher's Answers, but Its Internal Computations
Cognition Passes $1 Billion in Annualized Revenue Two Years After Launching Devin
Cursor Releases Agent SDK: Agents as Code, Deployed With a Single Command
Cognition Expands into Latin America: Devin Already Runs at Itaú and Nubank
Cognition opened an office in São Paulo. Over 75% of Itaú's tech teams use Devin, and Nubank cut a monolith migration from years to weeks at a 20x lower cost.
Healthcare Organizations Use Claude to Combat Ebola in DRC
A rare strain of Ebola is spreading in the DRC without a confirmed vaccine. Global healthcare organizations are deploying Claude to accelerate response to the…
Cognition launches Fusion in Devin: a two-model pairing cuts task costs by up to 39%
Cognition releases SWE-2: 50% on FrontierCode at a 64% lower price
Cognition Brings On the Dioxus Team, Keeping 4 Open Source Projects Under Devin's Wing
Cognition raises more than $2 billion at a $48 billion valuation as Devin revenue approaches $900 million
Anthropic in "legal limbo": two US courts can't agree on whether the military can use Claude
A US appeals court has contradicted a lower court's March ruling, putting the US military's right to use Anthropic's Claude model in question.
US Army builds its own chatbot, VICTOR, to support soldiers on missions
The Pentagon is building VICTOR, a military AI chatbot trained on real Army data, to deliver mission-critical information to soldiers in the field.
Project Maven: How the Pentagon's Skeptics Became Military AI's Biggest Believers
Project Maven, the Pentagon AI program launched in 2017, turned the defense establishment's most vocal skeptics into committed advocates of combat AI.
Pentagon told Anthropic it was in "near-complete agreement" a week before Trump announced the split
Palantir doubles down on military AI: takeaways from its developer conference
Pentagon labels Anthropic a national security threat over its AI ethics limits
