Wednesday, October 14, 2026
TL;DR
Character.AI filed for Chapter 11, selling assets to Disney for $20 million after years of lawsuits over teen safety. Anthropic needed its own Git replacement, Scion, after code changes passed 30,000 merges a day. Patch Visual Studio Copilot now.
Act on this
- Patch Visual Studio and SSMS: a Copilot path-check bypass let agents overwrite protected files like .mcp.json for code execution. Microsoft fixed it Oct 13. Rehberger, embracethered.com #
Signals
Verification is replacing code review, not improving it #
Geoffrey Huntley argued across five posts this week, Oct 7 to Oct 13, that hand-typing code is obsolete and human line-by-line review can't scale with agent output - closing with 'There's no place in the world now for a software engineer who types code by hand.' Simon Willison amplified Huntley's own pre-commit hook, which fails any commit that bypasses an approved diagnostics library, the same day. Huntley, agency · Willison, quoting Huntley
Two trackers read the same data point as evidence AI research is accelerating on its own - and one credible voice disputes it #
Jack Clark's Oct 12 Import AI moved his own recursive-self-improvement timeline from 2028 to 2027, citing TasteVal, a benchmark where Claude Opus 5.5 beat the human baseline on research judgment at roughly 1/30 the expert cost. Zvi Mowshowitz flagged the same Anthropic data independently as a possible RSI warning sign. Nathan Lambert pushed back Oct 9, calling the trend 'parallelized, AI-assisted language modeling' and much of the RL data behind it still 'remarkably low-quality.' Clark, Import AI 476 · Zvi, AI #189 · Lambert, Interconnects
News
WATCHAnthropic built its own Git replacement after AI pushed code volume past 30,000 merges a day #
Anthropic engineers shipped Scion, an internal source-control system, after charting merge volume against their own model release dates and finding standard Git tooling couldn't keep up with AI-driven commit volume. Anthropic Engineering
SHIPAnthropic previews Flows, letting a lead Claude agent write its own multi-agent coordination #
Instead of a fixed orchestrator config, a lead agent in Claude's new Flows writes the program that coordinates its subagents. Anthropic engineers say the fix for runaway parallelism is better evals and lower defaults, not more structure. Latent Space
WATCHJudge orders OpenAI to hand over hiring records in Apple trade-secrets case #
A federal judge ordered OpenAI to produce interview materials for former Apple hardware staff it hired, plus private messages from some, within two weeks. Apple's request for OpenAI's own R&D files was denied for now. TechCrunch
SHIPGitHub ships a new AI model for secret detection, cutting false positives #
GitHub's security team rebuilt its secret-scanning detector to catch more real credentials while cutting noisy false positives, folding validator and pattern work into one model across its default secret-scanning product. GitHub Blog
WATCHMatthew Green launches a site estimating total human effort spent on cryptanalysis #
Built using Claude and GPT-6 Astra, bitsofwork.org tries to estimate cumulative human-hours spent attacking the hard problems behind public-key cryptography, with Green flagging the methodology as rough and inviting correction via GitHub issues. Green, A Few Thoughts on Cryptographic Engineering
The long view
A year ago, Character.AI responded to mounting lawsuits over harm to minors by banning users under 18 from open-ended chat with its AI companions, rather than winding the product down. This week it filed for Chapter 11, selling what's left to Disney for $20 million plus a tech license, with no buyer yet for the rest of the company. If this holds, a company Google valued enough to license for $2.7 billion becomes a cautionary tale for the next wave of AI-companion startups chasing the same audience it couldn't keep safe or solvent.
Also noted
- Jesse Vincent's Superpowers plugin turns one, now Claude Code's top third-party plugin with 1.1M+ installs; ships v7.0 this week. Vincent, fsck.com #
- Cerebras CEO Andrew Feldman pushed back on AI "doomsday" framing at TechCrunch Disrupt, urging more focus on AI's upside. TechCrunch #
- Hamel Husain and Shreya Shankar opened the last 2026 AI-evals cohort; Shankar separately demoed an LLM-classification speedup technique. Husain · Shankar, Full Stack Data Lab #
- Andrej Karpathy reportedly described AI access as a "steeper funnel" - roughly 20 million people now use frontier agent workflows regularly. via Hacker News #