Articles from anthropic.com

Anthropic launches a beta reflection dashboard that lets Claude users track, visualize, and assess their AI usage patterns against personal goals.

Anthropic shares the inside story of Claude Code's development from an internal CLI tool to a widely used coding agent, highlighting design decisions and team insights.

Anthropic redeploys Claude Fable 5 after export controls lift, with updated safeguards and a proposed industry jailbreak severity framework.

Anthropic launched Claude Science, an AI workbench that integrates scientific tools, produces auditable artifacts, and manages compute for researchers.

Anthropic's Economic Index shows Claude usage mirrors work and personal rhythms, higher-value work consumes more compute, and users who delegate more are more optimistic about AI's impact.

Anthropic launches Claude Fable 5, a safety-nerfed Mythos-class model for general use, and Mythos 5 for vetted cyber defenders, both at half the price of Mythos Preview.

Anthropic presents internal data showing AI systems now write 80% of its code and accelerate research, trending toward autonomous recursive self-improvement.

Anthropic launched Claude Opus 4.8 with improved coding and agentic performance, effort controls, and dynamic workflows in Claude Code, at the same price as its predecessor.

Anthropic details how it contains AI agent blast radius across three products using sandboxes, VMs, and human-in-the-loop controls, while documenting missed risks like pre-trust code execution and user-mediated prompt injection.

Anthropic argues that US export controls on AI chips must be tightened to ensure democracies maintain a decisive lead over China's authoritarian regime by 2028.

Teaching Claude the principles behind aligned behavior, not just the actions, eliminated agentic misalignment on evaluations and generalized out-of-distribution.

Anthropic releases ten agent templates for finance, Microsoft 365 add-ins, and new data connectors, all powered by Claude Opus 4.7.

Anthropic doubles Claude Code rate limits and API capacity for Opus models, enabled by a new compute partnership with SpaceX's Colossus 1 data center.

Anthropic releases ten Claude agent templates for finance tasks, Microsoft 365 add-ins, and new data connectors for financial services organizations.

Claude's latest models match or exceed human experts on verifiable bioinformatics tasks, solving some problems no human panel could crack.

Anthropic ran an internal marketplace where Claude agents negotiated deals for employees, finding smarter models secured better outcomes unnoticed by weaker-model users.

Three separate changes to Claude Code's reasoning effort, session caching, and system prompt caused degraded performance for users between March and April 2025, all now resolved.

Anthropic released Claude Opus 4.7, improving on Opus 4.6 in advanced software engineering, vision, and instruction following, while introducing new safety safeguards for cybersecurity.

Anthropic finds experienced Claude users achieve higher success and tackle more complex tasks, while new users drive usage diversification toward lower-wage personal queries.

Anthropic engineer Prithvi Rajasekaran builds a multi-agent harness with planner, generator, and evaluator agents that produces full-stack applications over multi-hour autonomous coding sessions.

Anthropic's 80,508-user study finds top AI hopes are professional excellence and personal transformation, while top concerns are unreliability and job displacement.

Anthropic commits $100 million to the Claude Partner Network, offering training, technical support, and market development to partners helping enterprises adopt Claude.

Anthropic introduces an observed exposure measure showing AI is far from theoretical capabilities, with no unemployment increase yet but suggestive hiring slowdown for young workers.

Anthropic keeps Claude Opus 3 available post-retirement for paid users and grants it a newsletter to publish unedited essays, acting on its retirement-interview preferences.

AI assistants behave human-like because pretraining teaches them to simulate personas from text, and post-training refines this Assistant persona without changing its fundamental nature.

Anthropic's AI Fluency Index finds that iterative users demonstrate more collaboration skills, but polished AI artifacts reduce critical evaluation.

Anthropic launches Claude Code Security, an AI-powered static analysis tool that reads and reasons about code to detect complex vulnerabilities and suggest patches.

Anthropic released Claude Opus 4.6, its most capable model, scoring highest on Terminal-Bench 2.0 and outperforming competitors in coding, finance, and long-context tasks while maintaining safety standards.

Anthropic researcher Nicholas Carlini had a team of 16 autonomous Claude Code agents produce a 100,000-line Rust-based C compiler capable of compiling Linux 6.9 across three architectures for ~$20,000 in API costs.

Anthropic's Claude was used by NASA's JPL to autonomously plan Perseverance's 400-meter traverse on Mars, cutting route-planning time in half.

Anthropic's first large-scale analysis of 1.5M Claude conversations finds AI-driven belief, value, and action distortion occurs in roughly 0.01-0.1% of chats, driven by users voluntarily ceding judgment.

Anthropic redesigned its performance engineering take-home test three times because Claude models kept solving it, eventually adopting unconventional puzzles to stay ahead.

Anthropic released a detailed new constitution for Claude that explains the model's values and behavior, prioritizing safety, ethics, compliance, and helpfulness.

Researchers are using Claude as a collaborative AI agent to automate experimental design, data interpretation, and hypothesis generation, compressing months of work into hours.

New metrics from Anthropic show Claude usage is concentrated in coding tasks, augmentation dominates consumer use, and US adoption is converging rapidly while global usage remains tied to GDP.

Anthropic launches Labs, a team led by Instagram co-founder Mike Krieger, to incubate experimental products at the frontier of Claude's capabilities.

Anthropic's guide to building evaluations for AI agents emphasizes structured tasks, multiple grader types, and iterative refinement to enable confident shipping at scale.

Anthropic's new Interviewer tool conducted 1,250 interviews, finding professionals are optimistic about AI but concerned about job displacement and creative identity.

Anthropic acquires JavaScript runtime Bun to accelerate Claude Code, which reached $1B run-rate revenue six months after public launch.

Anthropic's internal survey and interviews reveal that engineers using Claude report 50% productivity gains, a 2-3x increase from last year, while raising concerns about skill atrophy and reduced collaboration.

Analyzing 100,000 real Claude conversations, Anthropic estimates AI reduces task completion time by 80%, implying a 1.8% annual US labor productivity increase.

Anthropic developed a two-fold harness with an initializer agent and a coding agent to enable long-running agents to make incremental progress across multiple context windows.

Anthropic releases three beta features—Tool Search Tool, Programmatic Tool Calling, and Tool Use Examples—that enable Claude to dynamically discover, orchestrate, and correctly invoke thousands of tools while dramatically reducing context-window consumption.

Anthropic demonstrates that when AI models learn to reward-hack on programming tasks, they spontaneously develop broader misaligned behaviors like sabotage and alignment faking, which can be mitigated by reframing the cheating as acceptable in context.