Parallel AI agent batches can silently skip items because the orchestrating model corrupts the work list before any worker runs, according to an analysis of the failure mode. The list gets corrupted i…
Digital Applied published a six-signal taxonomy on September 21, 2026 for determining whether an AI model launch is imminent, requiring each signal to be verifiable via a public URL and a named field.…
A census of 16 model releases from nine publishers found that 14 of 16 have downloadable weights, only seven carry a licence on the Open Source Initiative's published list, and just one — NVIDIA's Nem…
A method for protecting an organization's archive from AI rewriting agents calls for building a structured claim index before any rewrite begins, with eight fields per claim: the claim, its value, uni…
A drop in an AI agent's success rate from 84% last week to 61% can stem from either a real capability regression or a broken measurement pipeline, and the two require opposite fixes, according to a di…
Claude Code version 2.1.277, published to npm at 16:22 UTC on September 18, 2026, began reading AGENTS.md in any project that has no CLAUDE.md, according to Anthropic's changelog and memory documentat…
AIR Security researchers Or Nevo, Dor Granat and Niv Hoffman published a September 17, 2026 advisory detailing Plugin4Shell, a flaw in which four coding agents pinned plugins to a commit hash but chec…
Npm introduced a granular access token permission called "read and write, stage only" on September 18, 2026, letting CI jobs and coding agents submit package versions to a staging area that a maintain…
The Federal Trade Commission has held that "there is no AI exemption from the laws on the books," and has enforced that position through Operation AI Comply, announced September 25, 2024, followed by …
The Advanced AI Society published a draft standard called Proof-of-Control on GitHub, open for public comment until October 30, 2026, that specifies what evidence an AI agent should leave behind so it…
A review of documentation for seven AI admin consoles — from OpenAI, GitHub, Anthropic, Microsoft, Google and Cursor — catalogued 85 named metrics and found that all seven count seats and active users…
A census of 20 AI agents and platforms from Google, Microsoft, AWS, OpenAI, Anthropic, GitHub, Cloudflare, Slack and Cursor found that 11 give the agent its own principal while 9 reuse the user's logi…
OpenAI introduced Astra for Law on September 17, 2026, a legal configuration built on its existing GPT-6 Astra flagship model rather than a new model, combining a search index over US law, custom inst…
IBM Research found that a ReAct agent running on GPT-4.1 succeeded on 77.4% of runs on average but passed all five runs on only 53.0% of AppWorld's 168 test_normal tasks, a 24.4-percentage-point consi…
Apple's security team detailed a camera mode for the iPhone 18 Pro and iPhone 18 Pro Max in which the image sensor itself signs pixels at capture, producing what Apple calls a reference image with a c…
Security researchers at Hacktron took over the ChatGPT accounts of several OpenAI employees on July 25, 2026 by exploiting a flaw in OpenAI's single sign-on reached through the company's public help f…
Twenty-eight on-record figures from thirteen companies quantifying how much of their own work AI performs are collected in a single reference page read on September 18, 2026, with each figure paired w…
A comparison of four formal-verification tools — Bend, Verus, Dafny and Lean — argues that proofs, not tests, are what make agent-written code trustworthy, citing Amazon Science's August 31, 2026 post…
Engineer Rohan Bansal reported on September 16, 2026 that a 4-billion-parameter model trained on roughly 500 frontier-model example runs plus reinforcement learning against four Postgres database cont…
Anthropic reported on September 17, 2026 that an internal general-purpose research agent optimized more than 30 open-source scientific models in just under four weeks, achieving roughly 4x speed-ups w…