CrowdStrike's Cyber Superintelligence Lab found that the most advanced publicly deployed content safety classifier, which guards models including Claude Opus 5.5 and Fable 5, achieved a 0% direct bypa…
The European Commission designated ChatGPT a Very Large Online Search Engine under the Digital Services Act on Aug 31, 2026, citing its hybrid prompt-and-search nature and 159.1 million average monthl…
A developer examined whether Markdown bold in agent instruction files like CLAUDE.md, AGENTS.md, GEMINI.md and SKILL.md actually influences model behavior, concluding it does not. The analysis cites P…
A developer argues that adding a second AI model to review code changes does not improve trust in automated review, because models trained on similar data share the same blind spots and only re-read t…
PrismML released a ternary-weight version of Qwen3.8-27B under its Bonsai family on Tuesday, cutting the model from more than 50 GB of weights at full precision to just under 6 GB, which lets it run o…
An OpenAI researcher with over fifteen years of AI experience, including nearly five years at OpenAI, warns that increasingly situationally aware language models are eroding the ability to evaluate th…
A Digital Applied guide published September 12, 2026 argues that rising AI usage only shows more work is passing through an AI system, not that teams are completing more useful work, and proposes a de…
Emanuel Perez used OpenAI's GPT-6 Astra to build a Mac app that turns a MacBook's speakers and microphone into a contactless scrolling controller, emitting a 20 kHz tone and interpreting Doppler shift…
Microsoft Research researchers Minseon Kim, Zhengyan Shi and colleagues present FrogNano, a 4B coding agent post-trained exclusively with reinforcement learning on about 1,500 software engineering env…
Microsoft, through Tracer AI, has rescinded its DMCA copyright takedown notice against Luanti's Google Play listing, and the open-source voxel game app has been reinstated on the Play Store. The withd…
Microsoft Research's DELEGATE-52 benchmark study found that frontier AI models lose roughly 50% of document fidelity over extended multi-step workflows, with catastrophic corruption occurring in over …
Sylvain Kalache, who leads AI Labs at Rootly and was an SRE at LinkedIn, argues that AI incident-response tools create "comprehension debt" by auto-resolving the routine outages through which responde…
Microsoft Research released BitNet b1.58 2B4T, the first open-source native 1-bit LLM at the 2-billion parameter scale, which uses ternary weights (-1, 0, +1) requiring about 1.58 bits per weight, red…
Researchers at Microsoft Research, led by Nik Swamy, used AI coding agents to port and evolve a verified mark-and-sweep garbage collector for OCaml into a generational garbage collector, integrating i…
AI coding assistants like GitHub Copilot accelerate code generation, but verification of correctness, security, and safety has become the bottleneck in software delivery, according to an analysis by a…
Microsoft Research published a paper showing that a 4-billion-parameter AI model trained with SocialRL, a cascade reinforcement learning approach, achieved an average utility score of 0.627 across six…
GitHub Next released WikiKB, a knowledge base layer built on the GitHub repository wiki that enables humans and agents to ingest, search, and query curated knowledge efficiently. WikiKB leverages Self…
Microsoft Research's Webwright lets AI web agents write command-line programs instead of clicking through tasks, boosting the same GPT-5.4 model's success rate on long-horizon tasks from 33.5% to 60.1…
Recordari, a custom memory graph built by developer Corby, scored 0.831 on the LoCoMo benchmark, outperforming Microsoft Research's Memora, which published a score of 0.801. The benchmark, run with th…
Microsoft Research introduced CARE-X, a unified chest X-ray vision-language model designed to improve clinical usefulness by combining generative report generation with calibrated diagnostic predictio…