Recent stories
Each story's summary, with the places it was discussed.
Summaries are written by AI from the linked sources. Open the source for full context.
Claude Duet is a proactive agent that works in a cloud computer through connected apps, briefing users, completing tasks, and requesting approval for sensitive actions.
It shares Claude’s memory, skills, and connectors in one text or voice thread, producing documents, sheets, decks, and dashboards. Available as a research preview to US Claude Pro and Max subscribers on a waitlist, it shares plan usage limits and may misjudge urgency or miss newly arriving information.
Claude Fable 5.2 Scores 65 on Artificial Analysis, Highest So Far
Artificial Intelligence · Large Language Models · Anthropic · Claude · LLM Benchmarks
Claude Fable 5.2, Anthropic’s proprietary model released in November 2026, scored 65 on the Artificial Analysis Intelligence Index—the highest score reported so far.
The index combines nine evaluations covering knowledge work, automation, coding, scientific research, reasoning, document analysis, and physics. The model takes 171.16 seconds to produce its first answer token, generates 101 tokens per second, requires 45 minutes per index task, and costs an estimated $26.54 per task, with no public API available.
Anthropic’s Claude for Financial Services repository provides reusable Cowork plugins and Claude Managed Agent templates for investment banking, research, private equity, wealth management, fund administration, and onboarding.
It includes workflow agents, modeling skills, commands, MCP data connectors, Microsoft 365 deployment tooling, and human-review safeguards; firms can customize prompts, systems, templates, and compliance processes.
Anthropic's internal AI use, at retail prices, now exceeds its employee pay
Artificial Intelligence · Large Language Models · Business · Anthropic · Claude
Anthropic’s internal use of Claude, valued at the retail prices it charges customers, exceeded monthly employee compensation excluding stock awards in April, according to its IPO filing.
By September, that usage had risen more than 250-fold since January. The company says Claude wrote over 90% of code added to its software that month, while revenue grew much faster than headcount. The figures support treating AI tokens as a compensation component alongside salary, bonuses, and stock.
Anthropic is investigating elevated errors affecting authentication to Claude.ai and Claude Code.
The incident, posted Nov 13, 2026 at 09:58 UTC, also affects Claude Console (platform.claude.com) and Claude Cowork. The company has not yet identified the cause or provided a resolution, and says it will issue an update as soon as possible.
Early progress on accelerating drug development with Claude
Artificial Intelligence · Science · Health · Anthropic · Claude
Anthropic has begun expanding Claude’s role across drug development through partnerships with six organizations and its own programs, reporting early preclinical results.
Generate shortened a CAR T-cell genetic component by 70%, Divergence generated 35,000 antibody designs, BigHat improved lead properties, and Enveda cut document-review time 89%. BMS is evaluating clinical-trial plans, while Mithril research produced promising mouse results in under 18 months. Anthropic plans broader partnerships, clinical integration, and tools including Claude Science and research programs.
The SEC should block Anthropic’s IPO until the company provides reliable financial disclosures.
Although Anthropic reports $96 billion in cash, it has at least $582 billion in non-cancelable spending commitments and nearly $900 billion in total contractual obligations, while its 2026 results remain unaudited and revenue growth has slowed. Weak demand, intense competition, infrastructure costs, and data-center delays could make these obligations unmanageable, exposing workers, retail investors, retirees, and the financial system to substantial risk.
Anthropic has acquired Stainless, a specialist in SDK, CLI, and MCP server tooling.
This strategic move aims to enhance Claude's ability to act as an agent by improving its connectivity to external data and tools, thereby advancing developer experience and the ecosystem of integrations available for building autonomous agent workflows.
Anthropic has apologized for using invisible guardrails to throttle Claude Fable 5, a practice that hindered AI research and development.
The company will now make these safeguards transparent, routing distillation-related queries to Claude Opus 4.8 instead of silently degrading responses. This shift aims to restore trust after backlash regarding restricted access to frontier models.
Anthropic has confidentially submitted a draft S-1 registration statement to the SEC for a proposed IPO.
The move allows the company to explore going public subject to market conditions and regulatory review. Pricing and share details remain undetermined as the firm continues its expansion in the AI sector.
Following Anthropic's acquisition of Bun, concerns have arisen regarding the quality and direction of the toolchain.
While Bun remains a high-performance JavaScript runtime, the declining user experience of Claude Code, characterized by aggressive restrictions and billing issues, suggests potential 'enshittification.' Consequently, some developers are migrating projects to pnpm to mitigate reliance on Anthropic-owned infrastructure.
How the Claude Code team thinks about verification
Artificial Intelligence · Testing · Anthropic · Claude · AI Agents
The Claude Code team uses verification loops to let coding agents prove changes with evidence and scale parallel work.
It prioritizes end-to-end tests, property tests, and differential tests, with fast, reliable checks run before commits. Verification skills exercise real systems; failures escalate to humans after three rounds. Review rules stay outside the change, trigger from diffs, and are evaluated against incidents, reverts, and stable controls. Missed bugs become reviewed rules, while flaky checks are fixed or removed and verification systems pruned.
After seven months, Project Glasswing has produced 442 Anthropic-attributed CVEs, only two exploited in the wild, while Anthropic’s ledger lists 37,966 findings but 217 fixes.
Since September, findings rose by nearly 12,000, whereas fixes increased by 15; duplicates and withdrawals also grew, despite the claimed true-positive rate reaching 94.2%. Anthropic’s OSS Scanner shifts validation to maintainers: curl rejected 19 of 34 reports, confirmed six low-severity vulnerabilities, and left nine pending, challenging the project’s effectiveness and severity claims.
Researchers discovered the 'J-space' in Claude, a set of internal neural patterns acting as a 'workspace' for deliberate reasoning.
Emerging during training, this space functions similarly to human conscious access, allowing the model to hold thoughts, reason, and modulate information internally without explicitly writing it, providing insights into model behavior and hidden potential misalignments.
Claude for Foundation Models is a Swift package enabling Claude as a server-side model within Apple's Foundation Models framework.
It conforms to Apple's LanguageModel protocol, allowing developers to use standard APIs for streaming, tool calling, and structured outputs. The package facilitates seamless switching between on-device models and Claude, with requests routed directly to Anthropic.
Anthropic has partnered with SpaceX to access significant compute capacity at the Colossus 1 data center, enabling higher usage limits for Claude Code and the Claude API.
The company is actively expanding its infrastructure through strategic deals with Amazon, Google, Microsoft, and NVIDIA to support growing international demand and enterprise requirements.
Anthropic reports that newer Claude models require 80% less system prompting than their predecessors.
By moving from rigid rules to model judgment, using progressive disclosure, and leveraging rich references, users can improve AI agent performance. The core approach shifts from manually constraining agents to designing better interfaces, tools, and modular context structures.
Shipvideo is a serverless OpenComputer agent that turns URLs or prompts into deterministic MP4 films without manual editing.
A TypeScript agent uses Claude Opus 5.5 and three tools to fetch websites, validate scenes, and render frames through headless Chromium and ffmpeg in disposable arm64 microVMs. Deployable with one click, it is fully open in a GitHub repository.
Anthropic priced its initial public offering of Class A common stock at $83 per share.
The offering includes 980 million shares, with Anthropic selling 956,351,354 and certain stockholders selling 23,648,646; Anthropic receives no proceeds from the latter. Trading is expected to begin on Nasdaq Global Select Market under “ANT” on November 19, 2026, with closing expected November 20. Underwriters also have a 30-day option for 147 million additional shares.
Progress on provable inference for large language models
Artificial Intelligence · Cybersecurity · Large Language Models · Computer Science · Anthropic
Anthropic and collaborators released FlashProve, an open-source prototype for provable inference using zkSNARKs.
It commits to a model’s hidden weights and proves responses came from that committed model, allowing CPU verification without rerunning inference. On a single H100, it proves Llama 4 Maverick (400B-parameter MoE) at about 110 tokens/second, with 19 MB proofs per 1,024-token request—roughly 70–95× vLLM’s GPU time. The release includes C++ and Lean verifiers, formalized security claims, code, and documentation; it hides weights but not architecture and supports integer greedy decoding.