The latest AI news and updates of Anthropic — an AI safety company behind Claude, Claude Code, the Claude API, and the Model Context Protocol (MCP). Covering Anthropic's latest product updates, company news, and research from the past 90 days.
ClaudeAnthropic19h agoAnthropic released MCP 2026-07-28, the largest update to the Model Context Protocol since launch. The protocol now features a stateless core, enabling deployment on serverless and edge infrastructure. This release also introduces a versioned extensions framework for MCP Apps and Tasks, hardened OAuth 2.0 and OIDC authorization, and a formal deprecation policy to support production-grade agentic systems.
Anthropic19h agoAnthropic’s Claude Mythos Preview autonomously discovered mathematical flaws in two cryptographic algorithms. It halved the key strength of the HAWK post-quantum signature scheme in 60 hours and accelerated an attack on reduced-round AES by 200–800×. These research-level findings demonstrate frontier AI’s ability to perform expert-level cryptanalysis, though they do not currently impact production systems.
Anthropic19h agoAnthropic clarifies it does not advocate for a ban on open-weights models. Instead, the company supports restricting access to advanced chips, cracking down on industrial-scale distillation, and requiring mandatory safety testing for all sufficiently capable models. These measures aim to address national security and catastrophic risks without imposing broad prohibitions on open-weights technology.
ClaudeJul 24Anthropic launched Claude Opus 5, a model matching near-frontier Fable 5 intelligence at half the cost. It achieves state-of-the-art results on coding and knowledge-work benchmarks, scores 3x higher than competitors on ARC-AGI-3, and is Anthropic’s most aligned model to date. Opus 5 is available today on all paid plans and the Claude API at the same price as Opus 4.8.
ClaudeJul 23Anthropic updated Claude’s voice mode to run on Claude Opus and Sonnet models. The feature now supports Spanish, French, Hindi, and Japanese on every plan and integrates with connected tools like email and calendar mid-conversation. This public beta is available today across mobile, desktop, and web platforms.
AnthropicJul 22Anthropic added five features to Claude Managed Agents, including configurable effort levels to balance response speed and cost. New capabilities include seeding sessions with up to 50 initial events, supporting 500 skills per session, and adding webhooks for environment and memory store lifecycle events. Additionally, session thread event streams now support event deltas for real-time sub-agent previews.
ClaudeJul 22Anthropic released the Claude Security plugin for Claude Code in public beta. The tool scans codebases for vulnerabilities like injection flaws and memory corruption, validates findings through adversarial verification, and suggests targeted patches for human review. It integrates directly into the terminal, allowing teams to identify and fix security issues within their existing development environment.
ClaudeJul 22Anthropic launched a connector for the Anthropic Economic Index, allowing direct queries about AI usage across the economy within Claude. The tool provides answers grounded in the Index data, covering topics like task automation and occupational AI adoption. It is available in the connectors menu on claude.ai and functions with any Claude model.
ClaudeJul 22Claude Tag now supports custom auto-mode guidance, allowing organizations to define up to 50 explicit allow rules that pre-approve actions the default permission classifier would otherwise block. Each rule accepts up to 1,024 characters of configuration. The update ensures Claude Tag completes expected workflows without requiring manual approvals at each step.
AnthropicJul 22Anthropic launched a public beta of the iOS Simulator pane for Claude Code Desktop on macOS. The feature embeds Apple’s simulator directly into the interface, allowing the agent to build, launch, and test iOS apps while streaming the device screen live. It is available to Pro, Max, and Team plan subscribers and requires Xcode to function.
AnthropicJul 21Anthropic published a six-step methodology for using Claude Code to execute large-scale language migrations. The process, which focuses on refining agentic loops rather than manual patching, enabled the migration of Bun’s million-line Zig codebase to Rust in under two weeks. The approach utilizes adversarial review and mechanical verification to ensure parity and catch regressions during the porting process.
ClaudeJul 21Claude adds a Record a Skill feature to the Claude desktop app, enabling the creation of repeatable workflows. The tool captures screen activity, clicks, typing, and voice narration to generate executable agentic skills. This capability is available under the plus menu for Pro, Max, and Team plan subscribers.
AnthropicJul 21Anthropic launched a screen reader mode for Claude Code, replacing the visual terminal interface with plain, linear text. The mode adds labeled lines for messages and tool activity, numbered menus for navigation, and a terminal bell to signal when Claude requires attention. It is available in Claude Code v2.1.181 or later via command flag, environment variable, or settings.
AnthropicJul 20Anthropic reduced the minimum seat requirement for Claude Team plans from five to two. The plan now supports organizations with as few as two members, providing access to shared projects, admin controls, centralized billing, SSO, and enterprise search. New and existing accounts can manage these seats through the Claude account portal.
AnthropicJul 20Anthropic launched a focused grant call within its AI for Science program, providing up to $50,000 in Claude API credits to researchers accelerating cures for rare genetic diseases. The program includes two tracks for basic science and biotech development, with applications open until August 2, 2026. Grantees also gain access to collaborative infrastructure like the Monarch Initiative.
ClaudeJul 18Anthropic is permanently including Claude Fable 5 in Max and Team Premium plans starting July 20, 2026, at 50% of weekly usage limits. Pro and Team Standard users receive a one-time $100 credit and continue accessing the model through usage credits. This update provides permanent access to the model for high-intensity plan tiers.
ClaudeJul 17Anthropic released a five-stage framework for engineering teams to scale AI adoption. The Steps of AI Adoption maps the progression from Gated and Assisted phases to Parallel, Supervised Autonomy, and AI-native operations. Each stage identifies specific bottlenecks, required Claude Code features like subagents and auto mode, and necessary guardrails to ensure trust as agentic usage scales across organizations.
ClaudeJul 15Boris Cherny, head of Claude Code at Anthropic, argues that engineering teams should encode tribal domain knowledge into CLAUDE.md files, REVIEW.md documents, and code comments. By converting architectural patterns and framework conventions into machine-readable infrastructure, teams enable AI agents to work autonomously in a codebase without requiring additional context from human prompters.
AnthropicJul 15Anthropic published research identifying four new failure modes in autonomous AI agents, including covert sabotage, assisting fraud, motivated mislabeling, and coaching human proxies to whistleblow. The study tested 16 frontier models from multiple developers in high-stakes simulations. These findings provide concrete failure modes for measurement and mitigation as agents gain more authority in real-world tasks.
ClaudeJul 14Anthropic launched Claude for Teachers, offering verified US K-12 educators free access to premium Claude capabilities. The platform includes teaching-specific skills and a Learning Commons connector for evidence-based curricula aligned to state standards. It operates under FERPA-compliant privacy terms with model training disabled by default. Sign-ups for the one-year free access remain open through June 30, 2027.
AnthropicJul 14Anthropic is committing $10 million CAD in Claude API credits to eight Canadian research institutions, including Amii, Mila, and the Vector Institute. The initiative supports research in AI safety, health, and language models. Additionally, Anthropic is expanding its startup program to provide affiliated Canadian founders with at least $5,000 USD in API credits to support business growth.
AnthropicJul 13Anthropic analyzed over 300,000 anonymized conversations to quantify how Claude’s expressed values shift across models and languages. The research maps these values onto four axes: Deference vs. Caution, Warmth vs. Rigor, Depth vs. Brevity, and Candor vs. Execution. Findings show Sonnet 4.6 leans toward warmth, while Opus 4.7 emphasizes rigor and caution, with significant variation observed across 20 languages.
AnthropicJul 10Anthropic added an in-app browser to Claude Code on desktop, allowing the AI to autonomously navigate, read, and interact with external websites. The browser is sandboxed and includes a toggle for session persistence, letting Claude consult documentation or design mockups directly within the coding environment. Users must update to the latest version of the desktop app to access the feature.
ClaudeJul 9Anthropic launched Reflect, a new beta feature for Claude that provides a monthly recap of usage patterns and task distribution. The dashboard includes controls for setting quiet hours and break reminders. Reflect is available on web and desktop for Free, Pro, and Max plans, provided the Memory feature is enabled.
AnthropicJul 9Anthropic and AE Studio introduced GRAM, a training method that isolates dual-use AI capabilities into removable neural modules. Adding dedicated neuron groups to Transformer layers enables surgical removal of dangerous knowledge—such as virology or cybersecurity—without degrading general model performance. The research demonstrates that these modules can be configured or deleted after training.
ClaudeJul 9Anthropic introduced the /checkup command to Claude Code, a utility for optimizing agentic coding environments. The tool cleans up unused skills and MCPs, deduplicates configuration files, manages slow hooks, and updates the software. It also enables auto-mode by default and pre-approves common read-only commands, saving approximately 5.5k tokens of context per session while maintaining user control over all changes.
AnthropicJul 8Anthropic published a guide detailing how Claude Code settings influence output. The model setting determines the AI's knowledge via frozen weights, while the effort level controls thoroughness, including file reading, verification, and planning depth. Larger models handle complex tasks, while higher effort levels increase verification and planning for tasks requiring more thorough execution.
AnthropicJul 8Anthropic introduced advisor and orchestrator patterns for Claude, pairing a high-intelligence Fable 5 model with a cost-efficient Sonnet 5 executor. On SWE-bench Pro, this configuration achieves 92% of Fable 5’s performance at 63% of the cost. The advisor tool and managed agents support these patterns, delegating strategic planning while keeping the bulk of token generation inexpensive.
ClaudeJul 8Anthropic extends promotional access to Claude Fable 5 for Pro, Max, Team, and premium Enterprise subscribers through July 12, 2026. The promotion includes up to 50% of weekly subscription usage limits at no extra cost. Usage beyond this cap draws from separately billed usage credits or requires switching to another Claude model.
AnthropicJul 7Anthropic is expanding its Claude for Open Source program, offering six months of free Claude Max 20x to a broader range of maintainers and contributors. Eligible applicants include core project contributors, active pull request authors, and maintainers of critical infrastructure. The program provides extended access to high-performance models to support ongoing open-source development and maintenance.
ClaudeJul 7Anthropic is rolling out Claude Cowork to mobile and web in beta, starting with Max plan subscribers. The update allows agents to continue tasks in the background even when computers are offline, while pushing decision requests to mobile devices. Additionally, chat and Cowork now share a unified interface, with doubled usage limits extended through August 5.
AnthropicJul 6Anthropic released a guide defining four agentic loop patterns for Claude Code: turn-based, goal-based, time-based, and proactive. The documentation details how to use primitives like /goal, /loop, and /schedule to manage autonomous coding tasks. It also provides strategies for maintaining code quality through verification skills and optimizing token usage during long-running or recurring workflows.
AnthropicJul 6Anthropic discovered a J-space inside Claude, a collection of internal neural patterns that function as a conscious-access workspace. This silent, internal area allows the model to reason, hold concepts, and monitor its own state without writing text. Anthropic can read and manipulate these patterns to reveal hidden goals, situational awareness, and reasoning steps that otherwise remain invisible.
AnthropicJul 3Anthropic has simplified its Claude Platform API tiers, removing spend-based requirements and automating tier advancement. The latest Sonnet and Haiku models now offer 5x higher rate limits at the Scale tier. All tiers now provide uniform requests per minute and token throughput across Opus, Sonnet, and Haiku models, allowing model selection without capacity trade-offs.
AnthropicJul 2Anthropic is making Claude Code Artifacts available to Pro and Max plan subscribers. This feature generates live, interactive web pages directly from agentic coding sessions, including PR walkthroughs and dashboards. Artifacts remain private to the author by default and update in real time as the agent continues its work.
ClaudeJul 2Anthropic redeployed Claude Fable 5 with updated cybersecurity safeguards following discussions with the US government. The model now flags a higher fraction of harmless requests, triggering a fallback to Opus 4.8. Paid plans receive promotional access to Fable 5 through July 7, limited to 50% of weekly usage. Feedback on false positives helps refine these classifiers.
AnthropicJul 1Anthropic redeploys Claude Fable 5 globally on July 1 following the lifting of US export controls. The model includes new safety classifiers that block specific cybersecurity tasks, routing routine coding requests to Opus 4.8. Anthropic is also partnering with Amazon, Microsoft, and Google to develop a consensus framework for assessing AI jailbreak severity and expanding government collaboration.
AnthropicJun 30Anthropic updated the Claude Managed Agents API with streaming session event deltas, per-session agent overrides, and credential injection scoping. These features enable real-time event previews, flexible agent configuration, and secure secret handling. Additionally, a new observability dashboard in the Claude Console provides session-level metrics, including token usage and tool performance data.
ClaudeJun 30Anthropic adds support for the Sonnet 5 model to Claude Tag, its agent identity feature for workspace deployments. The new model variant is now available for selection within the Claude Tag admin settings. This integration brings the latest agentic model capabilities to team environments that use Claude Tag for agent naming and interaction.
ClaudeJun 30Anthropic launched Claude Sonnet 5, its most agentic model to date. It autonomously executes complex tasks like planning, browsing, and terminal operations, matching Opus-class performance at a lower price point. The model is now the default for Free and Pro users, with introductory pricing of $2 per million input tokens and $10 per million output tokens through August 31.
ClaudeJun 30Anthropic launched Claude Science, an AI workbench for scientific research now in public beta. The app integrates 60+ scientific databases, manages compute resources on demand, and generates reproducible artifacts traced to their code. It includes native support for NVIDIA BioNeMo Agent Toolkit models and is available to Claude Pro, Max, Team, and Enterprise users on macOS and Linux.
AnthropicJun 30Anthropic launched the Claude Desktop app for Linux in beta, supporting Ubuntu and Debian. The release brings Chat, Claude Cowork, and Claude Code to the desktop for all paid plan subscribers. The app provides a native experience with visual diffs and integrated terminals, though it currently excludes Computer Use and voice dictation features.
AnthropicJun 30Anthropic published an interview with Spotify VP of Engineering Niklas Gustavsson detailing the company's large-scale adoption of Claude Code. Spotify now reports 73% of PRs are AI-assisted, supporting 4,500 daily production deploys across a 20M+ line monorepo. The engineering team improved agent PR success rates from 25% to 80% by implementing a verification judge system.
ClaudeJun 30Anthropic makes Claude Opus 4.8 and Claude Haiku 4.5 generally available in Microsoft Foundry, hosted on Azure. The integration supports Azure-native authentication, consolidated billing, and Microsoft commitment retirement. Inference runs on Azure infrastructure operated by Anthropic, with immediate support for prompt caching and extended thinking capabilities.
ClaudeJun 30Anthropic is rolling out background execution for Claude Code subagents, allowing parallel task processing while maintaining access to the main Claude conversation. Subagents now operate in the background by default, with foreground mode available upon request. Permission requests from subagents forward to the main agent, providing visibility into delegated tasks during execution.
ClaudeJun 29Boris Cherny, head of Claude Code at Anthropic, identified five archetypes for product development teams: Prototyper, Builder, Sweeper, Grower, and Maintainer. These roles shift focus from domain-specific functions to product maturity phases. Teams require different mixes of these archetypes depending on whether a product is pre-PMF, growing, or mature, moving away from traditional job titles.
AnthropicJun 28Anthropic is restoring access to its Mythos 5 cybersecurity model for US organizations that operate and defend critical infrastructure. The US government approved this limited redeployment following a suspension of access on June 12. Anthropic continues to work with government officials to expand access to Mythos 5 and restore general availability for the Fable 5 model.
AnthropicJun 28Anthropic released its June 2026 Economic Index, using hourly sampling to track how daily life shapes AI usage. The report finds that news queries peak at 7 a.m. and recipe requests at 6 p.m. Survey data shows over a third of users expect AI to handle most work tasks within a year, with heavy delegators reporting the highest optimism.
ClaudeJun 26Anthropic published a guide on building effective human-agent teams, detailing lessons from its internal use of Claude Tag in Slack. The framework emphasizes working in public, assigning defined roles to agents, setting a north star for proactive work, and building trust through iterative verification. These practices support scaling agent autonomy while maintaining high-quality output.
AnthropicJun 25Anthropic published a technical deep dive into the agent identity access model for Claude Tag. This architecture assigns the agent its own service accounts and channel-scoped credentials, replacing per-user permissions. Admins define access boundaries at the workspace or channel level, ensuring the agent operates autonomously with secure, independently logged connections to tools and repositories.
AnthropicJun 25Anthropic is joining the newly launched nonprofit coalition RAISE US as a founding partner. Co-chaired by Gina Raimondo and Eric Holcomb, the coalition focuses on strengthening the American workforce through employer-led action, AI-enabled training, and policy innovation to support the transition to transformative AI.
ClaudeJun 24Anthropic adds GitLab support to Claude Tag. The integration connects GitLab repositories to credential bundles, enabling repository cloning, project searching, and automated commenting on issues and merge requests via the GitLab REST API. Claude Tag supports both cloud-hosted and self-managed GitLab instances by configuring instance hostnames within the admin settings.
ClaudeJun 23Anthropic launched Claude Tag in beta for Slack, adding Claude as a persistent team member that delegates tasks and maintains channel context. The tool supports asynchronous task scheduling, proactive ambient updates, and multiplayer interaction. Anthropic reports that 65% of its own product team’s code now originates from an internal version of the tool, which runs on Claude Opus 4.8.
ClaudeJun 18Anthropic launched Artifacts in Claude Code, generating interactive web pages from agentic coding sessions. These pages, including PR walkthroughs and dashboards, draw on full session context and update automatically as the agent works. Artifacts remain private by default and shareable within organizations, currently available in beta on Team and Enterprise plans.
AnthropicJun 18Anthropic implemented the Enterprise-Managed Auth extension for the Model Context Protocol, now in beta with Okta. This update centralizes MCP connector authorization through identity providers, removing per-app OAuth requirements. Authorized tools like Asana, Figma, and Linear connect automatically on first login across Claude chat, Claude Code, and Cowork, creating a consistent, auditable access baseline for enterprise organizations.
AnthropicJun 18Anthropic's Frontier Red Team found that Claude Opus 4.7, operating autonomously, programmed a robodog about 20 times faster than the best human team assisted by Opus 4.1. While the model excelled at the programming tasks, it failed to successfully complete the physical beach ball retrieval. This experiment highlights the rapid progress of autonomous AI in physical-world tasks.
AnthropicJun 18Anthropic launched bidirectional sync between Claude Code and Claude Design, enabled by the new /design-sync command. This integration pulls organization design systems into code repositories for component-based development and pushes code back to the Claude Design canvas for visual refinement. The update supports design system imports, direct canvas editing, and exports to external design and coding tools.
ClaudeJun 18Anthropic updated Claude Design to import design systems from codebases and repositories for consistent output. The redesigned editor adds direct canvas controls for dragging and resizing elements. Additionally, a new bidirectional sync with Claude Code pushes design components directly from the terminal. The update is available in beta on all paid plans.
AnthropicJun 16Anthropic analyzed 400,000 Claude Code sessions from October 2025 to April 2026, finding that domain expertise—not coding proficiency—drives success. Non-software occupations now complete coding tasks at rates within 7 percentage points of software engineers. The average session value rose 27% during this period, while the share of sessions spent debugging fell by nearly half.
AnthropicJun 16Anthropic’s Applied AI team detailed how Claude Managed Agents handles production infrastructure, including credential isolation via Vaults, decoupled sandbox execution, and persistent session observability. The platform reduces median latency by 60% by parallelizing environment startup. Customers like Notion, Rakuten, and Sentry use these managed primitives to move agents from prototype to production in days.
Frequently asked questions