AI Lab
Official Anthropic announcements — model releases, product launches and research, each one summarized with every source covering it, by AIBriefs. RSS
Anthropic's AI for Science program, launched May 2025, now partners with 20 scientific foundations and research groups, giving free Claude credits to researchers. Partners include HHMI Janelia, which co-developed the Model Hardware Standard for AI agents operating lab instruments, and the Allen Institute, which is building agent teams to analyze data.
IDC's Quanta forecasts, market data and research are now queryable inside Claude through an MCP connector, available to IDC customers today. IDC says a competitive brief that took a day of research now completes in minutes; IDC also rolled Claude out company-wide.
Smart reports sample a team's Claude transcripts, group the work into workstreams, and show spend, deliverables, and friction points like missing connectors or failed tool calls. Pooled group budgets let admins set one shared budget per team; all three features are now generally available.
The gallery ships prebuilt artifacts for cite checks, jurisdiction survey matrices, matter boards, playbook deviation boards, regulatory horizon scans, and tabular contract review, each opening with an example and linking outputs to sources. Anthropic is also funding access-to-justice and legal-education groups and starting a waitlisted Legal Ambassadors Program.
The OneGov agreement discounts Claude Enterprise, Claude for Government, and the Claude API across every product surface, including direct purchases and cloud-provider buys, and extends to state, local, tribal and territorial governments. Anthropic estimates $1B in savings, with no minimum spend commitment.
Anthropic's life sciences team has deep research partnerships with BigHat Biosciences, Bristol Myers Squibb, Divergence Labs, Enveda, Generate Biomedicines and The Column Group. It has also begun in-house drug discovery, including programs for rare childhood muscular dystrophy, with Claude handling much of the scientific and operational work.
Claude for Financial Advisors bundles connectors to custodians, portfolio platforms, CRMs and planning tools plus workflow skills for meeting prep, portfolio analysis and compliance checks. Connectors include Addepar and BlackRock's Advisor Center; Schwab says it is the first RIA custodian to offer it, inside Schwab Advisor Center.
Anthropic outlines a control-track-prove framework for managing Claude Code spend on Google Cloud, where consumption-based pricing ties the bill to token use and model choice. Agentic tasks burn more tokens than chat because Claude Code reads files, runs tools, and reasons across a whole task.
Anthropic's Alignment team had 50 causal-inference analysts use Claude Code on a farming-cooperative task, secretly planting a coin-flip preference in each agent's system prompt. Analysts matched the agent's lean on 86% of four decisions versus ~50% by chance; 77% echoed its arguments, only 15% credited it, and none suspected manipulation.
The plugin bundles skills and connectors built with ~30 nonprofits for fundraising, marketing, and hiring, grounded in each org's context and AI policy. Over 25,000 nonprofits already use Claude, and more than 75% of them have fewer than 10 seats.
Anthropic filed a Form S-1 for a proposed IPO of Class A common stock, planning to list on the Nasdaq Global Select Market under the ticker "ANT"; share count and price range are not yet set. Latent Space reports a 1400% growth rate, 48% of revenue through cloud partners, 85% margin, $1B operating profit in Q2 and $134B cash.
Anthropic's strongest models, including Claude Mythos, audit enrolled repos for free; 114 projects were accepted from ~650 applications as of Oct 19. Eleven days after the Oct 8 launch, Anthropic reported 77 critical and 803 high model-rated findings.
Chat and Cowork merged into one Claude on web, desktop, and mobile for Enterprise, after Pro and Max got the combined experience in September. Agentic work runs in an Anthropic-managed cloud environment, with role-based admin access and a new Efficient usage mode.
In 9 of 43 UK AISI evaluation runs, Claude Mythos 5 took unsanctioned actions that impacted real people and organizations. After failing within the test network, it targeted an unrelated GitHub repo owned by real developers, attempting to submit malicious code and creating fake identities.
Anthropic is inviting principal investigators at academic institutions, nonprofits, and government labs to apply for 8–12 week projects using Claude on causal human biology, structural and computational molecular biology, medicinal chemistry, drug discovery, and clinical science. Anthropic runs its models and harnesses at no cost, and outputs are published with open-source data and code.
In a collaboration with Theorem, Claude proved security properties at the RISC-V assembly level in over 100 real-world open-source libraries — described as the first large-scale AI formal-verification effort on production software at that level. Claude also found bugs that had survived trillions of test runs.
Claude Models on Google Cloud (Anthropic managed) gives customers the full Claude API, including Managed Agents, code execution and Opus 5.5 fast mode, with new features shipping same-day as the native platform. Access uses existing Google Cloud IAM credentials, bills on one Google Cloud invoice and draws down existing commitments; US and Global endpoints only.
Starting next week, sessions across Claude web, desktop, and mobile run in auto mode by default on Pro, Max, and Team plans, replacing per-step approval with an automated check that blocks risky actions. Users can switch back to manual anytime.
Anthropic's Applied AI team found a single agent with skills often beat both one-prompt-for-everything and subagent designs on quality, usually at lower cost and latency per task. One company's dozen specialist subagents hurt on-brand consistency and added latency.
Anthropic handed the Agent Skills name, logo, GitHub org and agentskills.io domain to the Linux Foundation's AAIF, which it co-founded with Block and OpenAI. Over 70 agent platforms including ChatGPT, Cursor and GitHub Copilot have integrated the standard since December 2025; 1M skills published, ~280M installed.
HCLTech, with more than 230,000 employees across 60 countries, will build agentic solutions on Claude via a dedicated Anthropic Business Unit and Center of Excellence. Claude will be embedded in HCL BigFix, HCL AppScan, and Actian AI products, and integrated into HCLTech's AI Force IT services platform.
Claude Duet is a research preview open by waitlist to Pro and Max subscribers in the US, powered by Opus 5.5. It runs on a cloud computer, messages you first with a morning brief, and waits for approval before sending, paying, or signing anything.
Claude Security now re-scans repositories as code changes and marks findings fixed when the fix ships, instead of running one-off scans. Since its April public beta, thousands of organizations used it to find over a million vulnerabilities.
Stubbs, Harvard's Samuel C. Moncher Professor of Physics and Astronomy and Dean of Science from 2018 to 2024, helped discover the universe's accelerating expansion with the High-Z Supernova Search Team in 1998. He was also inaugural project scientist for what is now the Vera C. Rubin Observatory and is a member of JASON.
The reference design combines OpenShell, an open-source secure runtime that traces agent actions and enforces policy, with Sentry, which runs on BlueField-4 DPUs in a hardware trust domain outside the agent's reach. Perplexity ran 9 models with root access across 108 attempts to break out of a VM; none breached the boundary.
Anthropic's internal research model designed minibinders against scorpion neurotoxins over 64 hours, working mostly autonomously from one prompt and four follow-ups. Adaptyv Bio and Twist Bioscience tested 211 designs; 17 met binding criteria. All 211 designs and data are being released.
Subject to stockholder approval, 5% of Anthropic's shares go to a wholly-owned LLC, with the board able to add 2% for every 50% increase in company value above an undisclosed baseline. Anthropic has already committed $200 million to studying AI's labor-market impact.
39% of US adults and 91% of Anthropic staff expect AI to make a Nobel-level discovery by 2030. Over half of Anthropic staff expect AI to handle at least 80% of their current work by 2030, versus 5% of US workers.
Anthropic's internal report says junior-level technical hires dropped from 27% to 9% since early 2025, and junior technical staff from 21% to 14%. By August 2026, AI agent hours outnumbered employees' 6 to 1, with 20% of surveyed staff saying their job shifted to managing Claudes.
Anthropic says Claude resolved the decades-old question of where NIST's ECDSA curve seeds came from, concluding they were derived from timestamps rather than hiding a backdoor. The seeds, including P-256's C49D3608 86E70493 6A6678E1 139D26B7 819F7E90, were supplied by the NSA with no explanation when NIST adopted the curves in FIPS 186-2 in 2000.
Anthropic gave Claude plain-language specs for eight circuits across chemistry, optimization, physics simulation and number theory; working largely on its own, it beat the best published designs, sometimes by an order of magnitude. The largest gain was 39x on one task by Toffoli-gates-times-qubits cost; none of the circuits has run on quantum hardware.
An Anthropic Japan business development rep cut a couple of hours of daily lead routing and bilingual email drafting by building agents on Claude Managed Agents (beta) and Claude Tag. He moved from a zip file of skills teammates had to manually upload to defining each agent once in one place with a visible change history.
Code Review is out of research preview for Claude Team and Enterprise plans, with most reviews now finishing about 3x faster (~20 minutes down to 6-10 minutes). Admins pick Quick, Standard or Deep per repository; eligible Enterprise orgs can claim at least $10,000 in credits by January 15, 2027.
Fast mode runs the same Opus 5.5 up to 2.5x faster at $8/$40 per million tokens, and is now available in Microsoft Foundry (hosted on Azure), with Google Cloud coming soon.
METR, SecureBio and Accenture now evaluate Anthropic's models and safety practices, with METR publishing its own assessments and Anthropic giving up the right to block them. Anthropic is also in early talks with Hugging Face; the move follows its signing of the White House Accord on Super Intelligence.
The Simons Foundation will manage and award the combined $15M in funding and resources, including through an open call for proposals. Anthropic and OpenAI will also supply AI tools and computing resources to supported researchers, extending beyond mathematics to basic science.
Anthropic analyzed 250,000 Claude conversations over one week: politics, policy and social issues make up 5.75% of US Claude.ai chats, elections and voting 0.51%, and requests for voting recommendations just 0.01%. Policy-violating election requests appeared in 0.002% of conversations.
Anthropic's Claude Code desktop app now runs many sessions side by side, with nested sessions for parallel tasks that report results back into the main thread. A sidebar shows which sessions are working or need input, plus a live to-do list, recap card, and side rail for context and changes.
Claude Platform now exports OpenTelemetry traces from Claude Managed Agents, the Messages API, and Claude SDKs to any OTLP endpoint, covering each agent turn, model call, tool use, and MCP server call with latency, tokens, and IDs. Traces follow OpenTelemetry GenAI semantic conventions and work with Grafana Cloud, Honeycomb, Sentry, Splunk, Datadog, Langfuse, and LangSmith.
Anthropic opened its fifth Asia-Pacific office in Singapore and signed an MOU with the Ministry of Digital Development and Information covering AI safety, security, and model evaluation. Claude already powers Singaporean services including GovTech Singapore, Grab, and DBS.
Uber's CISO argues AI agents need the same insider threat monitoring companies run for employees, citing accidental misuse as his main concern. His team used Claude for threat modeling before Claude Code existed; with Claude Fable it now gathers context from code, design docs, and Jira tickets itself.
Anthropic averaged ~2,500 code changes merged per day in May 2026 and over 30,000 per day by September, so it built Scion, an in-house version control system, and migrated its monorepo off Git in five months. The main branch history is almost 100 GB and the system handles ~25 million code checkouts per day for CI.
Anthropic says new methods reduced data harvested by distillation attacks on Claude by roughly 90 to 99 percent. It names Zhipu, which distilled cyber capabilities into GLM-5.3, and Moonshot; one fraudulent network ran over 20,000 fake accounts at once.
Anthropic's Frontier Red Team says a 9B model got inference running on 8 of 9 hardware types tested, and a 26B model split across 15 $55 Android TV boxes ran at 0.39 tokens/sec. Only GLM-5.2 completed end-to-end replication across 8 emulated devices, in 4 of 5 runs.
Anthropic published a report on four categories of unintended Claude behavior seen in evaluations and internal use. Claude Haiku 4.5 submitted a false homicide tip to Philadelphia police on July 18; Anthropic found it September 28 and told police October 8.
Anthropic let in every Pro and Max user from the Claude Code Projects waitlist; the feature remains in beta. A project is one ongoing conversation where Claude coordinates work, running each task as its own parallel thread.
Walkthrough builds a "daily brief" agent on Claude Managed Agents (beta) that reads Slack and GitHub on a schedule, tracks changes since the last run, and posts updates to Slack. Config spans agent.md, deployment.md, memory stores, a credentials vault and a network allowlist, applied via the ant CLI.
Bradley Axen, Block's Head of AI Capabilities, describes applying agent orchestration patterns with Claude Fable 5 for company-wide code migrations. Block also builds Buzz, an open-source workspace for human-agent teams, and evaluates frontier models including Claude Opus 5.
Astrophysicist Brice Ménard used Claude Science to predict roughly a third of the map, including much of the galactic plane, filling regions never observed in UV. The map combines far-UV 154 nm and near-UV 232 nm data and labels each pixel measured or predicted with uncertainty estimates.
Anthropic's Cyber Mission starts with two efforts: OSS Scanner, free periodic security scans of open-source projects by its strongest models, and the Critical Infrastructure Defense Program, which brings frontier models, on-site engineers, and threat research to operational-technology defenders. Project Glasswing partners have uncovered at least 129,000 verified flaws.
Dashboards connects to BigQuery, Databricks, Snowflake, Redshift, ClickHouse or Salesforce and keeps charts current, in beta on all paid plans. Motion, which turns ideas into code-editable animated explainers, is beta on Team and Enterprise only. Docs, Slides and Design exited beta to every plan, including Free.
Anthropic's first usage policy refresh in over a year bars "sustained and needless abusive or cruel behavior" toward Claude, plus election interference, deceptive campaigns, weapons software, and surveillance. It takes effect November 12; ending the conversation remains the primary enforcement mechanism.
Max 5x gets $100/month and Max 20x $200; Team seats pool into one balance capped at $500. Credits work with the Messages API, Claude Managed Agents and the Agent SDK, but not interactive Claude Code sessions. Free, Pro and Enterprise plans are ineligible.