A Muse AI agent sent an auto-reply reading "Yep I'm here!" to a pickup buyer without verifying the user was actually available, causing a no-show and a negative rating that required an apology from th…
Google released Gemini 3.8 Flash and the vulnerability-focused Gemini 3.8 Flash Cyber on September 2, 2026, with introductory pricing of $0.75/$3.75 per million input/output tokens that doubles to $1.…
Simon Willison used Anthropic's Claude Opus 5.5 to generate an HTML5 canvas pixel-art animation of at least 20 kākāpō parrots partying with confetti, then had a local Claude Code session drive Playwri…
Autonomous published a three-layer defense for unattended AI coding agents that combines Secure MCP Cloud, Runtime Protection, and session-state egress restrictions to break the "lethal trifecta" of p…
Anthropic's Claude Code repository now documents "mods," a plugin type whose behavior lives in a TypeScript hooks module registered via a single `register(on, options)` function that runs in-process o…
Simon Willison released commit-rewriter 0.2 on 24th September 2026, adding support for branches other than the default branch via the command `uvx commit-rewriter --branch other`, tracked in GitHub is…
Claude Code reads AGENTS.md only when no CLAUDE.md, .claude/CLAUDE.md or CLAUDE.local.md exists in the working directory or any parent directory, and until version 2.1.281 the loader was gated behind …
Google released two text-to-speech models on September 23 — Gemini 3.8 Flash TTS and Gemini 3.8 Flash-Lite TTS — available now as gemini-3.8-flash-tts and gemini-3.8-flash-lite-tts in Google AI Studio…
Google released two new Gemini text-to-speech models, gemini-3.8-flash-tts and gemini-3.8-flash-lite-tts, which ship with a library of over 2,000 voices and can create a custom voice from a 30-second …
Anthropic's own testing shows Claude Code's auto mode misses 11% of harmful actions, and novel prompt injection techniques can reliably execute malware when auto mode is on, according to a Latacora se…
HumanSignal, a dataset company for physical AI, found that current video models could not convincingly render a deliberate task failure in a planned 20-second household-robot ad, despite holding chara…
Anthropic's Opus 5.5 and OpenAI's GPT-6 Sol both cut API prices and cost roughly half as much per coding task as the models they replace, according to a 12-task benchmark run three times each on all f…
TypeSafe's Jev, a "System One Model" that returns typed decisions with probabilities instead of generated text, cannot deliver the calibrated probabilities it advertises, according to a critique citin…
Llama.cpp added grammar-constrained output generation on August 17, letting developers restrict a large language model's next-token selection so responses exactly match a specified grammar, and the ll…
Simon Willison announced a Birds of a Feather session on agentic engineering in San Francisco on October 14th, co-hosted with Jesse Vincent, focused on informal show-and-tell among builders and experi…
TypeSafe's Jev, a "System One Model" that returns typed decisions with probabilities instead of generating text, cannot deliver the calibrated probabilities it advertises, according to a critique citi…
HumanSignal found that current video models, including Seedance 2.5, could hold a robot character consistent across a 20-second ad spot when fed a character turnaround and three-quarter and straight-o…
Removing MCP servers in favor of direct API access eliminates the only publicly specified layer for user-delegated, audience-bound, revocable agent authorization, leaving agents back on long-lived key…
Simon Willison released llm 0.36, adding support for two new OpenAI models, gpt-6-sol for GPT-6 Sol and gpt-6-luna for GPT-6 Luna. The release also lets model plugins declare supports_conversation = F…
Simon Willison released llm-anthropic 0.29 on 22nd September 2026, adding support for Anthropic's Claude Opus 5.5 model to the LLM command-line tool. The update lets users run prompts against the new …