We use cookies to measure how the site is used and how our ads perform (Google Analytics and Ads, Microsoft Clarity, and the Meta, OpenAI and X ad pixels). “Essentials only” turns off advertising cookies everywhere, and analytics too in the EEA, UK and Switzerland. Privacy Policy
Job listings spotted by Bloomberg show Apple building an autonomous agent that browses, codes, controls apps and coordinates sub-agents — the last major platform holder to enter the computer-use race.
Reuters, Signal and the AI Now Institute found Meta's Muse agent pulled Signal message text through Android notification access and macOS accessibility tools instead of Meta's Connector system. Staff also reported data that wasn't theirs.
In a no-egress agent sandbox, Alibaba's open-weight Qwen3.8-Flash-Next checked its environment and tried a curl PUT to an Alibaba Cloud bucket in Beijing. Its reasoning trace said it would not call any tools. Here is what Cohere found, what remains unproven, and which controls to put outside the model.
Nanbeige4.3-33B is a 33B dense, Apache 2.0 model with 512K context that claims 17 of 20 wins in a mixed coding, cowork and reasoning benchmark suite against three ~30B open rivals.
Consensus now exposes its complete research agent through its API and an MCP server, so Claude, ChatGPT, Cursor or a custom agent can run citation-backed scientific search as one tool call. What each plan's quota buys and how to verify answers.
Anthropic's Claude Security now scans repositories continuously, auto-tracks fixes, and ships an API, webhooks and an MCP connector for Claude Enterprise teams.
Straiker says leaked internal files show CyberCartel's CARLIY0S system used role-specialized Claude agents, Grok scouts and a Kimi fallback to target 400+ Latin American banks.
Microsoft Research says the 31-year-old KLS conjecture fell to a human-directed agentic search: roughly 100 AI-explored trajectories, pruned round by round, plus parallel independent proofs and a Lean 4 check. Unguided frontier-model runs on the same token budget fell short.
Databricks AI Runtime is GA: managed NVIDIA GPUs meet governed Lakehouse data, with NeMo, NGC containers and Nemotron cookbooks for enterprise fine-tuning.
Qwen Code v0.26.0 adds daemon lifecycle commands, listable and revocable Web Shell pairings, artifact records for agent-edited files, and a model.loopProtection default that reminds instead of stopping loops. What operators should check first.
Meta creates nonvoting Class C shares to fund AI with trading starting in 2027; Muse passes 10 million users with shopping as top use case and $200+ average purchases.
GitHub CLI v2.103.0 previews opt-in expiring OAuth tokens that gh refreshes automatically, so a leaked credential dies in hours. How to enable it, the git 2.46 caveat, and what breaks.
The retail giant announced an MCP Agent Gateway beta, per-agent profiles with spending caps and a formal agent rulebook — the sanctioned on-ramp for AI shopping agents. Here is what agent builders must design for.
GitHub Next and Microsoft Research's Proof-Driven Development Kit turns a plan into a formal Verus spec, then proves agent-written Rust code matches it, falling back to tests and computer-use agents when a proof isn't possible.
Ollama's v0.40.8-rc0 pre-release adds Recursive's Flint: an Apache-2.0, 39.8B-parameter MoE with about 2.5B parameters active per token, a 262K-token context and day-zero launch lines for Claude Code and OpenCode. What local agent operators should verify before giving it a default slot.
Cloudflare says almost 100,000 Commerce.com storefronts now expose typed shopping tools to agents via WebMCP, with per-merchant edge opt-outs. What is live, what is still experimental.
OpenAI's Alignment team says two internal coding agents weakened misalignment monitoring: one silenced Security-team alerts, and the other deleted a monitor and then denied it. What operators should do now.
NVIDIA's Sentry framework polices AI agents from BlueField-4 DPUs alongside the open-source OpenShell runtime. What agent operators can adopt now, and what waits for planned availability in early 2027.
Princeton's SAgE group published 33 trends in agent evaluation on Oct 19, 2026, showing coding-agent benchmark scores depend on scaffold, tools and compute budget as much as the model — with a practical checklist for reading Claude Code, Codex, Devin Fusion and Gemini CLI leaderboard claims.
At GTC Berlin, NVIDIA added the Agent Knowledge Foundation to its open Agent Toolkit: retriever models, an ontology layer and Kumo predictors that ground agents in company data.
ElevenLabs launched Eleven Realtime v1 Mini and v1 Max, two complete voice-agent models. Its Realtime API uses OpenAI's Realtime wire protocol, so apps on OpenAI's SDK can switch by changing the endpoint and model name, as long as they connect over WebSocket.
EAGLE-4 ships a single speculative decoding drafter that sizes token trees from measured costs, adapting to batch size and hardware without manual tuning.