Claude Frontier Academy: $100M to Train 10,000 Engineers
Anthropic announced October 2 that it is committing $100 million to train 10,000 Frontier Deployed Engineers by the end of 2027. The FDE Residency borrows from medicine: a multi-day in-person kickoff, then 12 weeks leading a real Claude project at your own organization.
First cohorts include Accenture, Bain, Capgemini, Commonwealth Bank of Australia, Deloitte, McKinsey, Morgan Stanley and Novo Nordisk, in San Francisco, New York and London. Credentials start in early 2027.
Claude Security Scans Enterprise Codebases
Anthropic released Claude Security, a tool that scans enterprise code for vulnerabilities and recommends fixes. A multi-stage system validates findings and assigns confidence scores so teams know what to fix first.
It adds scheduled scans and audit exports, with CrowdStrike, Palo Alto Networks and Infosys as partners. Enterprise customers only for now.
Fable 5.1 and Mythos 5.1: Same Model, Different Guardrails
Anthropic's September release gives general users Fable 5.1 and vetted security and life-science teams Mythos 5.1, which has fewer restrictions. Pricing is $10 in and $50 out per million tokens, with cache reads cut 75% to $0.25.
Anthropic says safeguards block 60% fewer false positives in cyber contexts. Mythos access runs through verification programs, US organizations only for now.
India Endpoint Keeps Claude Requests In-Country
Anthropic now processes Claude requests on servers inside India via Amazon Bedrock, a big deal for banks and government workloads with residency rules. Opus 5, Sonnet 5 and Haiku 4.5 are available through local inference.
India is Anthropic's second-largest market for Claude.ai, and software work makes up 45.2% of tasks there, the highest anywhere.
Releasebot Shows Mods as the Latest Claude Code Change
The most recent entry in the public release feed is still the October 1 mods launch: TypeScript functions that rewrite prompts, add UI and replace built-in features in the CLI and desktop. No new headline release landed over the weekend.
Mods run with full machine access, so only install ones you trust.
Anthropic Asks Users to Share Voice Data for Training
Claude users are now seeing a prompt: allow Anthropic to use voice conversations to improve its models. It is optional, with Allow and Not now buttons, and a separate Privacy toggle that is off by default.
You can switch it off or delete voice data any time in settings. It is separate from the existing chat and coding training options.
Anthropic Reports a Claude Threat to Florida Police
A Lee County, Florida woman was charged with a felony after telling Claude she would “shoot up” the sheriff's office on September 26. Anthropic's monitoring flagged it, human reviewers escalated, and police were notified.
Tom's Hardware says it is at least the third such chatbot case reported to police since August. Anthropic's privacy policy allows disclosure to prevent serious harm.
Anthropic Is Betting on People and Places, Not Just Models
A $100M engineer academy, in-country inference in India and a security product with named partners all say the same thing: the bottleneck is deployment, not model quality. Anthropic is building the humans and the plumbing around the model.
The voice data prompt and the police report are the flip side. As Claude gets embedded everywhere, questions about what Anthropic sees, keeps and shares get louder.