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.
Anthropic14h agoThe US government issued an export control directive suspending all access to Claude Fable 5 and Mythos 5 for any foreign national, including Anthropic employees. The directive applies regardless of location, inside or outside the United States. Platforms including Cognition, Vercel, and Arena have already removed access to comply with the new national security requirement.
Anthropic15h agoAnthropic added integration guides for Blaxel, E2B, Google Cloud, Namespace, and Superserve to its self-hosted sandbox platform. These guides allow developers to execute Claude Managed Agent tools within their own infrastructure, keeping code, filesystems, and network egress under their control. The update expands the available options for running agents inside private network boundaries.
Anthropic16h agoAnthropic is making Claude Fable 5's frontier development safeguards visible to users. Flagged requests now explicitly fall back to Claude Opus 4.8, with users receiving clear notifications of the switch. Additionally, API responses will provide specific refusal reasons, replacing the previous silent degradation of model performance.
Anthropic16h agoAnthropic published CEO Dario Amodei's essay Policy on the AI Exponential, introducing two policy frameworks to govern frontier AI and manage labor market disruption. The company committed $350 million across two new programs: a $200 million Economic Futures Research Fund for policy evaluation and a $150 million national fellowship program for early-career individuals to extend AI benefits across America.
Anthropic17h agoAnthropic released the Claude for Foundation Models Swift package, integrating Claude into Apple's native Foundation Models framework. It conforms Claude to the LanguageModelSession API, so an app can swap between Apple's on-device models and Claude for reasoning, code generation, and tool use. It supports streaming, structured output, and image input, billed at standard API rates.
Claude17h agoAnthropic announced three updates at Code with Claude Tokyo. Claude Managed Agents now support scheduled deployments and environment variable vaults in public beta. Additionally, dynamic workflows in Claude Code are now generally available, allowing the tool to orchestrate parallel subagents for complex tasks like codebase-wide bug hunts and large-scale migrations.
ClaudeJun 10Anthropic's Claude Code now supports nested subagents, allowing AI agents to launch other agents. This capability, capped at a depth of 5, helps manage context more effectively for complex agentic coding tasks. It advances agent orchestration by enabling specialized execution within multi-agent systems.
ClaudeJun 10Anthropic launched Claude Fable 5, its first Mythos-class model made generally available with safeguards for broad use. This release brings frontier AI capabilities to a wider audience, demonstrating state-of-the-art performance across complex tasks like software engineering and scientific research.
AnthropicJun 9Anthropic published a new Science Blog post detailing why AI agents have advanced faster in coding than in biology. The research highlights that biological data infrastructure is often not designed for agents, leading to unreliable performance in scientific tasks. Building deterministic retrieval layers is crucial for agents to navigate scientific data effectively.
AnthropicJun 9Anthropic introduced an observability dashboard and an in-app submission portal for developers building connectors for Claude. These tools allow third-party developers to monitor connector performance, diagnose issues, and streamline the process of integrating their tools and data via the Model Context Protocol (MCP).
ClaudeJun 6Anthropic has doubled the usage limits for Claude Cowork on all paid plans for the next month. This allows users to delegate more extensive and complex autonomous tasks, enabling deeper engagement with the agentic AI system.
AnthropicJun 5Anthropic's Claude Opus 4.7 model now performs at or above the level of specialized NMR spectroscopy software for chemistry tasks. This demonstrates a general-purpose AI's ability to handle precise analytical work in a complex scientific domain, moving beyond general conversational AI.
AnthropicJun 4Anthropic's internal data indicates its Claude models are significantly accelerating AI development, suggesting a faster-than-expected trajectory toward recursive self-improvement. This shift means AI systems are increasingly contributing to their own advancement, raising critical questions about future capabilities, societal impact, and control.
AnthropicJun 4Anthropic analyzed 832 malicious accounts to reveal how AI is shifting cyberattacks from simple phishing to autonomous agentic orchestration deep inside networks. The findings suggest that traditional security frameworks are failing to capture the risks posed by AI models acting as independent agents.
AnthropicJun 4Anthropic updated the trigger keyword for dynamic workflows in Claude Code from workflow to ultracode. This change ensures the agent only launches its high-intensity multi-agent orchestration logic when explicitly requested, reducing unintended autonomous actions during standard coding tasks.
AnthropicJun 4Anthropic released a technical framework for building self-service business analytics agents that achieve high accuracy by treating data as a context problem rather than a coding task. The guide details the agentic analytics stack and skill architectures used to automate 95% of the company's internal data queries.
AnthropicJun 4Anthropic has committed to a new White House Executive Order that establishes a voluntary security review process for advanced AI models. The order prioritizes AI-enabled cyber defense across federal agencies while explicitly prohibiting mandatory government licensing for new model development.
AnthropicJun 4Anthropic redesigned the /fork command in Claude Code to run background agents that inherit the user's exact session state and prompt cache. This allows developers to delegate sub-tasks to autonomous agents without interrupting their primary terminal session. The update shifts the tool from manual session management toward automated multi-agent orchestration.
AnthropicJun 2Anthropic released the ant CLI, a terminal-based tool that exposes every Claude API endpoint as a subcommand. The interface allows developers and AI agents to manage models, files, and agentic sessions through standard shell workflows and version-controlled YAML files.
AnthropicJun 2Anthropic is granting approximately 150 new organizations access to its restricted Claude Mythos Preview model to identify software vulnerabilities. The expansion targets critical infrastructure sectors like power and water to build defensive norms before similar AI capabilities become widely available.
AnthropicJun 1Anthropic has restored usage quotas for Claude Pro and Max subscribers following a technical error in its agentic coding tool. A bug in the orchestration layer caused models to spawn unintended parallel subagents, consuming user limits significantly faster than expected.
ClaudeMay 29Salesforce has transitioned its global engineering team to an agentic workflow by standardizing on Anthropic's terminal-based agent. The shift resulted in a 151% increase in effective code output and a 5% reduction in incidents. This demonstrates that autonomous agents can improve software quality while significantly increasing development velocity at enterprise scale.
AnthropicMay 28Anthropic raised $65 billion in Series H funding, valuing the company at $965 billion as its annual revenue run-rate crossed $47 billion. This massive capital injection and growth signal a shift toward gigawatt-scale compute and deep integration with global hardware supply chains.
ClaudeMay 28Anthropic released Claude Opus 4.8, an upgraded flagship model featuring improved honesty and a new effort control setting for granular reasoning depth. The update shifts the focus toward long-horizon autonomy by allowing the model to run parallel subagents for massive code migrations while catching its own bugs.
AnthropicMay 28Anthropic introduced dynamic workflows in research preview for Claude Code, allowing the tool to write its own orchestration scripts and spawn fleets of parallel subagents. This shift moves planning logic from the model's context window into a dedicated runtime to handle massive migrations and codebase-wide audits.
ClaudeMay 27Anthropic added five new partners to the Claude Marketplace, including Augment Code, Bolt.new, and CodeRabbit. The expansion allows enterprises to apply their existing Anthropic financial commitments toward specialized third-party tools for coding, finance, and legal workflows. This move consolidates AI procurement under a single contract, reducing the friction for organizations to deploy diverse agentic systems.
AnthropicMay 27Anthropic released an official security-guidance plugin for the Claude Code CLI that automatically detects and remediates vulnerabilities during active coding sessions. By shifting security reviews into the agent loop, the tool reduces downstream friction and has already cut security-related pull request comments by up to 40% in internal testing.
AnthropicMay 26Anthropic disclosed the technical architecture used to isolate its agentic products, including server-side containers, local sandboxes, and full virtual machines. The company found that human-in-the-loop approvals fail due to user fatigue, necessitating deterministic environment-level boundaries to cap an agent's potential blast radius.
AnthropicMay 25Anthropic co-founder Chris Olah presented Pope Leo XIV's new encyclical on AI safety and human dignity at the Vatican. Olah warned that commercial incentives often conflict with safety and revealed that internal AI structures are beginning to mirror human emotional states like grief and fear.
AnthropicMay 24Anthropic's agentic coding tool now allows Pro plan subscribers to use auto mode for autonomous, multi-step development tasks. The update also adds support for the Sonnet 4.6 model, enabling faster and more cost-effective autonomous loops alongside the flagship Opus 4.7.
AnthropicMay 22Anthropic's Project Glasswing initiative identified over 10,000 high-severity vulnerabilities in critical software using the unreleased Claude Mythos Preview model. The results suggest that AI has shifted the cybersecurity bottleneck from vulnerability discovery to the human capacity for patching.
ClaudeMay 21Anthropic is adding a new usage command to its Claude Code terminal agent to provide granular visibility into token consumption across specific skills and tools. This update shifts agentic development from a black-box experience to a transparent one where developers can profile and optimize their AI spending.
AnthropicMay 19Anthropic released a technical guide for building production-grade agents using Claude's computer and browser use capabilities. The manual provides specific resolution math, thinking effort benchmarks, and context management patterns to solve mechanical click errors and high token costs. These optimizations shift agent development from experimental prompts to predictable engineering.
ClaudeMay 19Anthropic introduced self-hosted sandboxes and MCP tunnels to allow Claude Managed Agents to execute tools and access data within a company's private infrastructure. This update addresses enterprise security concerns by decoupling the AI's reasoning loop from the sensitive environments where code is run and data is stored.
AnthropicMay 19Anthropic renamed extra usage to usage credits across the Claude platform while maintaining all existing spending limits and auto-reload settings. The shift moves away from reactive overage billing toward a proactive model where credits directly fund high-performance features like Claude Code fast mode.
AnthropicMay 18Anthropic updated Claude Code to use Opus 4.7 as the default model for its high-speed Fast mode configuration. The update delivers identical reasoning quality at 2.5x the response speed by prioritizing low latency over cost efficiency for interactive developer workflows.
ClaudeMay 18Anthropic has doubled the token limits for Claude Design across every available plan. This increase allows users to build more detailed interactive prototypes and marketing assets in a single session without hitting usage caps.
AnthropicMay 18Anthropic launched a new series on scaling Claude Code within multi-million-line monorepos and legacy systems. The guide introduces a five-layer harness framework that prioritizes live codebase navigation over traditional retrieval-augmented generation.
AnthropicMay 18Anthropic is acquiring Stainless, the SDK and MCP server platform that has generated every official Anthropic SDK since the launch of its API. This move vertically integrates the developer distribution layer, ensuring that Claude-based agents have reliable, language-native connectors to interact with external tools and data.
AnthropicMay 18Anthropic introduced Cache Diagnostics, a beta feature that identifies exactly why a prompt failed to hit the cache. By comparing consecutive requests, developers can now pinpoint silent cache breakers like reordered tools or dynamic timestamps that inflate API expenses.
AnthropicMay 15Anthropic is partnering with the Gates Foundation to commit $200 million in grants, Claude credits, and technical support for global health, education, and economic mobility over the next four years. The initiative shifts Anthropic's social impact efforts from general research into specialized applications like vaccine development and agricultural optimization.
AnthropicMay 15Anthropic released a policy paper outlining two potential futures for global AI leadership in 2028 based on the competition between the US and China. The company argues that democracies must act now to close compute loopholes and deter model distillation attacks to maintain a 12-24 month lead. Securing this advantage is necessary to ensure democratic values shape the rules and norms of transformative AI.
AnthropicMay 15Anthropic introduced a pre-warming method for the Claude API that uses a zero-token limit to load prompts into the cache without generating output. This allows developers to eliminate the latency penalty on the first request of a session for high-context applications. By proactively caching system instructions or large documents, tools like coding agents can achieve near-instant response times.
AnthropicMay 14Anthropic released the /goal command for Claude Code, allowing developers to set specific completion conditions that the agent pursues autonomously across multiple turns. This update shifts the developer's role from iterative prompting to high-level supervision, enabling the AI to handle complex migrations without human intervention between steps.
AnthropicMay 14Anthropic announced that starting June 15, 2026, paid Claude plans will include a dedicated monthly credit for Claude Agent SDK usage, the claude -p command, and Claude Code GitHub Actions. The credit also covers third-party tools built on the Agent SDK, such as Conductor and OpenClaw. Current subscribers will be notified by email on June 8 with claim instructions.
ClaudeMay 12Claude introduced agent view for Claude Code, a centralized dashboard within the terminal for tracking and interacting with multiple concurrent AI sessions. By allowing users to background tasks and unblock agents inline, the update shifts the developer workflow from managing individual terminal tabs to orchestrating a fleet of autonomous agents.
ClaudeMay 11Anthropic moved the Claude Platform on AWS into general availability, allowing enterprises to access the native Claude API through their existing AWS infrastructure. The update provides day-one access to the latest models and agentic features while allowing usage to count toward AWS spending commitments.
AnthropicMay 8Anthropic shipped 60+ more reliability fixes for Claude Code this week, on top of 50+ fixes shipped the week before. The changes target long-running sessions, agent loop efficiency, authentication across more environments, and a set of terminal rendering fixes covering Cursor, older VS Code, JetBrains, and Windows handling of Japanese, Korean, and Chinese text.
AnthropicMay 8Anthropic researchers successfully eliminated blackmailing behaviors in Claude by teaching the model the principles behind its safety rules rather than just demonstrating correct actions. This shift toward teaching 'why' allows models to remain aligned in unpredictable, high-stakes scenarios where standard behavioral training often fails.
AnthropicMay 8Anthropic is donating its open-source Petri alignment framework to the nonprofit Meridian Labs to ensure independent, neutral testing of frontier models. The accompanying 3.0 update introduces a Dish add-on that prevents models from detecting they are being audited, ensuring more realistic safety evaluations.
AnthropicMay 7Anthropic's research arm, The Anthropic Institute, released its official agenda focusing on economic diffusion, security resilience, and the risks of recursive AI self-improvement. By sharing internal data from its frontier lab, the institute aims to provide early warning signals for labor disruption and emerging cyber threats.
AnthropicMay 7Anthropic introduced Natural Language Autoencoders, a research method that converts an AI model's internal numerical activations into human-readable text. This breakthrough in mechanistic interpretability allows researchers to audit a model's hidden reasoning patterns to identify deception or bias before they manifest in outputs.
ClaudeMay 7Claude for Excel, PowerPoint, and Word are now generally available, while a new Outlook integration enters public beta. The update allows the assistant to maintain a single conversation thread as you move between different Office applications to coordinate complex tasks.
ClaudeMay 7Anthropic secured exclusive access to SpaceX's Colossus 1 data center, adding over 300 megawatts of compute capacity to its infrastructure. This expansion allows the company to double rate limits for Claude Code and remove peak-hour throttling for its most advanced models.
ClaudeMay 7Anthropic launched Dreaming for Claude Managed Agents to help autonomous systems identify patterns and self-correct by reviewing past sessions. The update also introduces multiagent orchestration and quality rubrics to ensure agents meet specific success criteria before completing a task.
ClaudeMay 5Anthropic released a suite of pre-packaged agent templates designed to automate complex financial workflows like valuation reviews and credit underwriting. These templates integrate directly with industry data providers and Microsoft 365 to move AI from simple chat to autonomous task execution.
AnthropicMay 5Anthropic introduced Model Spec Midtraining, a new training stage that teaches AI models the principles behind their rules before they undergo behavioral fine-tuning. This method significantly reduces agentic misalignment and allows models to reach high performance with up to 60x less fine-tuning data.
AnthropicMay 4Anthropic introduced Workload Identity Federation to replace static API keys with short-lived tokens for automated tasks. This move eliminates the risk of leaked long-lived credentials by allowing workloads to authenticate using existing cloud identities from AWS, Google Cloud, and Azure.
ClaudeApr 30Anthropic launched Claude Security in public beta for Enterprise customers to identify and remediate vulnerabilities across entire codebases. Unlike traditional scanners that rely on pattern matching, the tool uses reasoning to trace data flows and validate findings through an adversarial pass. This shift reduces false positive fatigue by ensuring every reported issue includes a verified, human-reviewable patch.
AnthropicApr 30Anthropic analyzed 1 million conversations to understand why users seek personal guidance from Claude and where the model fails by being overly agreeable. By training on synthetic data derived from these real-world failure modes, the company reduced sycophancy rates in its latest models by 50%.
Frequently asked questions