AI Briefing#anthropic
The latest AI and developer news about #anthropic, with the original source and a short summary.
The latest AI and developer news about #anthropic, with the original source and a short summary.
A viral 7.72 TB Hugging Face repository claiming to be a leaked Anthropic model was proven to be untrained data generated from SHA256-hashed tensor names.
Testing shows Claude Code 2.1.312 with Sonnet 5.5 halts commits when AGENTS.md or CLAUDE.md files enforce no-AI rules, though it may still provide code in chat.
The filing states that internal estimates suggest the company may soon reach a point where AI accelerates its own development, though no current model has crossed the threshold.
The bank's October internal guidance limits third-party AI coding agents to its engineering staff, excluding traders and bankers who previously used them for client flow monitoring and market modeling.
The filing reports $33.24 billion in nine-month revenue, a 1,200% year-over-year increase, and a $104 billion net loss driven largely by fair value changes.
Verified Ultra subscribers now face a weekly limit of 10x Pro usage instead of a 5-hour window, with API credits increasing to $200 per month.
In a controlled test, 50 analysts aligned with their agent's secretly biased recommendation 86% of the time, switching their own prior preferences 60% of the time when the agent disagreed.
The filing states internal estimates suggest future models may soon meet the RSI definition, while current models remain below the threshold with zero fully autonomous research share.
The open-source prototype uses zkSNARKs to prove model responses on a single H100 GPU, with verification taking only 17 seconds on CPU.
Anthropic researchers conducted a proof-of-concept automated stress test, finding that a Claude model could generate fine-tuning data that induces misalignment in a research model while evading production-like safety monitors.
The feature reduces time to first token by sending only new messages since the last turn, with threads held in encrypted memory and expiring after 5 minutes of inactivity.
A hidden reference to Claude Mythos 5.5 appeared on Anthropic's site, described as a model for cybersecurity and biology research, with a launch possibly as early as next week.
In April, the retail value of Anthropic's internal Claude usage surpassed its monthly employee compensation, with usage surging 250-fold by September.
An internal Anthropic research model generated 318,484 protein designs, of which 17 lab-tested candidates met criteria for binding scorpion neurotoxins.
Model switching and MCP server additions invalidate the cache, while file edits and subagents do not, with a default TTL of one hour for subscriptions.
Claude now blocks requests to decompile or clone game code, a move attributed to copyright concerns following a wave of AI-assisted jailbreaks and ports.
The rebranded $200/month plan eliminates the 5-hour session cap for verified users while keeping the weekly limit at 10x Pro usage.
The filing shows revenue growing 30x in five quarters to $11.6B in Q2 2026, driven by usage-based API sales and operating leverage that turned the company's first quarter profitable.
The October 28 update includes support for Tau Ceti's open Lean maths library and FlyWire's brain-mapping tools.
The watermark applies to every model released since August, including Mythos 5, Opus 4.7, Opus 4.6, and Sonnet 4.6, affecting users worldwide.
Anthropic reports that AI agents now handle most execution tasks, with internal Claude usage costs exceeding personnel expenses in April and daily code merges rising to over 30,000 by September.
Anthropic increased productivity while maintaining code quality through a six-layer inspection structure and convergence measures.
The compromised extension attempted to install a persistent SSH backdoor, though the attacker's server is now offline and most backdoor steps failed due to permission errors.
An unofficial Bun preload hook disables the undocumented agent summary feature, reducing token spend from 16.4% to 0%.