All articles tagged with #claude
CoreWeave and Anthropic announced a multi-year agreement to run Anthropic’s Claude models on CoreWeave’s AI cloud, enabling production-scale workloads with high performance and reliability; the deal expands CoreWeave’s ecosystem to include nine of the top AI model providers and will begin rolling out later this year.
Bluesky unveils Attie, an AI assistant powered by Anthropic's Claude built on the AT Protocol to let users create custom feeds via natural language. Currently in closed beta as a standalone app, with plans to integrate Attie feeds into Bluesky and other atproto apps and, eventually, allow users to vibe-code apps atop AT Protocol.
Bluesky is launching Attie, a standalone AI app built by Jay Graber’s new team that uses Anthropic’s Claude to tailor feeds across Bluesky and other atpro networks, aiming to offer a more personalized, people-focused social experience. The project, described at Bluesky’s Atmosphere conference, marks a shift under new Bluesky CEO Toni Schneider, with Attie entering beta testing even as the platform remains wary of AI chatter among its user base.
Anthropic has tightened Claude's 5-hour session limits during peak hours (5 a.m.–11 a.m. PT) for free, Pro, and Max subscribers, while weekly quotas remain unchanged; about 7% of users may hit the new limits, and those with token-heavy workloads are advised to shift tasks to off-peak times as demand grows and the company scales its infrastructure.
Anthropic is throttling Claude during peak hours (5am–11am PT / 1pm–7pm GMT), redistributing its 5-hour weekly session limits and advising users to run heavy tasks off-peak; the move aims to manage demand. On the same day, OpenAI announced unlimited Codex access by removing caps to boost experimentation with new plugins, a stance likely temporary before monetization. The two shifts illustrate a broader, cyclical pricing/usage game among AI providers as users seek consistent access to models.
A federal judge laid out the core issues in Anthropic’s bid to block the Pentagon from labeling it a national-security risk, criticizing government actions as potentially punitive and raising questions about First Amendment rights, due process, and Congressional authority in a case that could reshape AI procurement.
A Tom’s Guide feature tests Claude’s Cowork and Dispatch features, which let Claude control and run tasks on your computer from your phone. Dispatch lets you issue tasks from mobile that execute on your desktop, requiring Claude Pro/Marx and the latest Claude Desktop, and it can interact with files, apps and services. Setup involves enabling computer use with appropriate permissions, keeping the computer awake, and reviewing results for accuracy. It signals a shift toward AI-driven delegation, saving time, but it’s early, not always perfect, and depends on user setup.
Insiders say the Pentagon views Anthropic’s Claude as highly capable for defense and cyber tasks, but a conflict over lawful-use constraints has stalled talks. A path to revive the deal could exist if negotiators agree on language that ensures lawful use and possible policy concessions, though any agreement would require intense mediation between Pentagon leadership and Anthropic’s CEO and board.
Anthropic unveiled a research preview that lets Claude control a Mac—clicking, typing, and navigating apps—through Claude Cowork and Claude Code, with Dispatch now extending to mobile for ongoing desktop tasks. The agent prioritizes service connectors, falls back to Chrome when needed, and only uses screen-interaction as a last resort, requiring per-app permissions and exposing screen content. Early tests show roughly 50% success on complex, multi-app workflows, and Anthropic is pricing for Pro/Max while pursuing enterprise plugins and governance features as it competes in the AI-agent race.
Anthropic’s Claude now includes agentic AI that can remotely operate a user’s macOS computer to perform repetitive tasks—like batch-resizing photos in Photoshop and adding logos—via the Claude Dispatch app, with previews limited to Claude Pro/Max; it prioritizes connectors to services but can execute tasks directly if needed, raising security concerns about accessing sensitive files.
A U.S. district judge questioned three Trump-era moves against AI maker Anthropic—the ban, Pentagon contractor cutoffs, and its supply-chain designation—saying they don’t clearly match the stated security risk and urging a return to the pre-designation status as Anthropic pursues court relief; the case, which includes First Amendment and procurement claims, continues with Anthropic seeking to pause the designation while litigation proceeds, and a decision was sought by March 26 though not binding.
Anthrop ic has added a research-preview feature that lets Claude autonomously perform tasks on macOS by connecting to apps and tools, with user permission and a manual fallback via keyboard and mouse. The capability, integrated with Dispatch for mobile task orchestration, is available to Claude Pro/Max subscribers on Macs and is designed with safeguards, but carries security risks like prompt injections and potential vulnerabilities; some apps are disabled by default and not all tasks work perfectly yet.
Anthropic’s Claude user survey of over 80,000 people across 159 countries finds AI hallucinations (mistakes) are the top worry for 27% of respondents, more than concerns about job losses (22%), with 16% fearing reduced ability to think critically. The study shows Claude is used as a work tool, educational resource, and personal aide, with 32% reporting increased productivity. Regional differences emerge: optimism is higher in parts of Africa, South America, and Asia, while Western countries show more concern about the economy. Researchers plan further targeted studies via Claude Interviewer, though experts caution about potential biases and methodological limits in the findings.
Anthropic says it cannot disable or alter Claude or provide a backdoor during military operations, disputing DoD claims it could sabotage the AI; the dispute centers on supply-chain risk and licensing, with court filings and hearings shaping how Claude may be used by the Pentagon.
The Trump administration, in a California federal court filing, defends the Pentagon’s designation of Anthropic as a national-security supply-chain risk after the AI firm refused to loosen guardrails on Claude. Anthropic is suing to block the blacklist, arguing violations of free speech and due process, while the government says the dispute is about contracts and national security, not speech. The filing suggests Anthropic’s ability to win may hinge on federal contracting implications and national-security concerns, with potential losses for the company and impacted government relationships.