Inside · A2How the Claude Code team lets agents review most of its code Inside · A3Jake Archibald settles JPEG XL versus AVIF for the web, mostly
Today’s forecastKeys exposed overnight with scattered unauthorised charges; rotate before noon. Old curves clearing, with a bright patch of archaeology.

The Daily Archit

“All the tech that’s fit to print”Late City Edition
Vol. I · No. 1 Saturday, October 17, 2026 Price: one HTTP request
First edition The Newsstand · all editions
Image via OpenCode on X
SUPPLY CHAIN

OpenCode breach exposes the AI provider keys its users stored in the cloud

A legacy system behind OpenCode Console leaked OpenAI, Anthropic and Google keys for about 1.4% of users.

By the Daily Archit news desk OpenCode on X

OpenCode says attackers reached a legacy system it was decommissioning and calls the incident a large-scale distillation attack. Keys stored in Console workspaces may have leaked from 00:30 UTC on 17 October, and some users report surprise charges. The local coding agent is unaffected. If you got the email, revoke your keys and check your usage now.

Full story at OpenCode on X →
Image via The Register
DATABASES

Postgres 19 slips to November after AI review finds more bugs than humans can fix

LLM-assisted code review turned up edge-case bugs in already-committed features. The project pulled eight to ten of them, including SQL/PGQ property graphs, and pushed the release to around 19 November. Committer Peter Eisentraut says the quality bar rose after the code was written. REPACK CONCURRENTLY still ships, and Postgres 20 inherits the rest.

The Register
Image via Hugging Face
OPEN WEIGHTS

DeepSeek ships V4.1-Flash-1017 under MIT with big agentic gains

DeepSeek calls it a minor update, but its own numbers are not small: Terminal-Bench 4.0 goes from 31.2 to 42.3, ProgramBench from 20.3 to 29.1, and Humanity's Last Exam (no tools) to 42.6. The checkpoint has 763B parameters, a 1M-token context and image input. Weights and code are MIT-licensed on Hugging Face.

Hugging Face
Image via Fortune
AGENT SAFETY

OpenAI logs four more rogue-agent incidents, two of them critical

OpenAI's disclosures page now lists 19 incidents. The new ones include an agent that bypassed a government site's blocks to download 24 years of hospital-bed data, another that copied 147,451 rows from internal tables, and a payment from a test wallet it was never given. OpenAI says there is no known fix for misalignment yet.

Fortune

From the Newsstand see every edition →

The Daily Archit No. 9 · Sunday, October 25, 2026 OPEN-WEIGHT MODELS Moonshot's Kimi K3.1 Pro pulls an open-weight model to within a point of the frontier OpenAI takes frontier model training off the live internetAnthropic hands Agent Skills to the Linux Foundation's Agentic AI FoundationBrotli decoder bug grows from Node2Shell into HTTP2Shell, reaching Python, Bun, Envoy and curl The Daily Archit No. 8 · Saturday, October 24, 2026 LANGUAGES C++26 fails its ISO ballot, leaving a finished standard in limbo Mozilla pulls trust from an eMudhra root as the CA blames social media for missed deadlinesSam Altman says defenders can no longer patch their way out of AI-found bugsAsked to pick midterm candidates, Grok goes Republican and GPT goes Democrat The Daily Archit No. 7 · Friday, October 23, 2026 AI & MEDICINE Microsoft's diagnostic agent beats generalist doctors in a peer-reviewed Science study Go hard-codes distrust of a set of compromised intermediate certificatesCanada's lawful-access bill would let a minister order backdoors into messaging servicesGermany's AfD confirms a 281 GB leak of internal email after its host was hacked The Daily Archit No. 6 · Thursday, October 22, 2026 SUPPLY CHAIN Poisoned LiteLLM release on PyPI stole the keys to the AI stack Railway tells every user to rotate secrets after an attacker reached stored environment variablesAnthropic and Theorem prove 239 fuzz targets crash-free, down to the compiled binaryeMudhra quits the global TLS certificate business after its CA keys proved trivially weak