Latest signal November 22, 2026 · afternoon edition
Sunday's AI signal turns on who does the measuring, as OpenAI published its own pretraining loss as a proposed safety yardstick, an outside group had to estimate Anthropic's 2028 power mix, and a Ghostscript exploit list that reaches AI image tools still has no public full exploit.
01 / The wire
Recent briefings
-
November 22, 2026 · afternoon
Sunday's AI signal turns on who does the measuring, as OpenAI published its own pretraining loss as a proposed safety yardstick, an outside group had to estimate Anthropic's 2028 power mix, and a Ghostscript exploit list that reaches AI image tools still has no public full exploit.
-
November 22, 2026 · morning
Zero Data Retention turned out to be a label rather than a guarantee, as the Dragos canary study, Together AI's leak disclosure and OpenRouter's removal of 125 endpoints from its ZDR list show prompts escaping at labs, resellers and hosts, including through a credential a coding agent committed.
-
November 21, 2026 · afternoon
A 17-million-line Lean proof that no person can read was certified because independent checkers agreed, and several other items today point the same way, output is cheap and the checker is the product.
-
November 21, 2026 · morning
OpenAI and Anthropic are fitting coding agents for work nobody watches live, through remote sessions, managed runners and a regression queue, and OpenAI's own write-up shows the trust sits in identities, disprove steps and a human merge, while Signal shows apps starting to block the agents that do watch the screen.
-
November 20, 2026 · afternoon
The newest agent gains come from fitting the harness rather than swapping the model, as Hugging Face trained small adapters to specific coding harnesses, Microsoft Foundry let browser agents run a whole click sequence as one script, and GitHub let team maintainers set default agent plugins for their teams.
-
November 20, 2026 · morning
Agent governance is collapsing into single chokepoints (one identity per agent, one MCP URL per tool kit, one switch per blueprint), and a chokepoint only protects what cannot route around it.
-
November 19, 2026 · afternoon
On November 18 the control points of an agent moved out of the prompt and into the platform, as Microsoft put versioned tools behind one MCP endpoint and outbound traffic behind fail-closed rules while LangSmith tied coding-agent spend to the pull request it produced.
-
November 19, 2026 · morning
On November 18 agents moved from asking to staying connected, as GitHub put event streams into gh and added eight MCP connectors and Claude Code let MCP-backed policy hooks block on failure, so what happens when a check breaks is now the setting to audit.
02 / Under the surface
Latest analysis
-
Zero Data Retention Isn't a Guarantee. Test Your AI Provider With a Fake Key
Zero Data Retention is a claim a provider makes about its own storage, and after the Dragos canary study builders should treat it as…
-
Upload2Shell: A Fake PNG Can Wake Up Ghostscript on Your AI Image Server
The Upload2Shell Ghostscript bug reaches self-hosted AI image tools because Pillow hands PostScript bytes inside a file named .png to…
-
unkernel Gives Your AI Agent a Disposable Linux Machine in Under a Second
Unkernel makes a disposable real Linux machine a single npx command or a JavaScript function call, which suits scratch environments for…
-
Fast-Anything Turns Your Coding Agent Into a Rust Surgeon, Not a Rewrite Crew
Fast-anything shows the safe way to let a coding agent speed up an app is to let a profiler pick one function, give the agent that function…
-
The Trending Blender AI Add-on Has No Code in Its Repo. Check This Before You Install Any.
The Blender AI add-on climbing GitHub's trending boards ships no source code in its repo, only a Windows ZIP from an account a week old, so…
-
Rustrak Puts Your AI Agent Traces on Your Own Server. Check Three Settings First.
Rustrak lets you keep AI agent traces, which Sentry's AI integration fills with full prompts and outputs by default, on a box you own with…
-
OpenAI's Performance Factory Turns Codex Into an On-Call Team
OpenAI's Performance Factory shows that the useful unit of agent automation is a small state machine with the human steps written into its…
-
AI Models All Give the Same Answer. Two Prompts Make It Yours.
Models from different labs converge on the same default answers, a habit added during fine-tuning, so anything built on a model's default…
04 / Coverage map
Topics we track
Claude Code 128 Anthropic 53 OpenAI 45 GitHub Copilot 27 Codex 25 Agent Skills 19 Hugging Face 19 MCP 17 LangChain 16 Cloudflare 15 DeepSeek Harness 14 Model Context Protocol 13 Kimi K3 11 METR 11 Microsoft Foundry 9 Anthropic Frontier Red Team 8 GPT-5.6 Sol 8 Claude Code auto mode 7 Codex CLI 7 GPT-6 Astra 7 MCP 2026-07-28 7 Claude Fable 5.1 6 Cursor 6 GPT-5.6-Cyber 6