Anthropic opens up its most powerful AI yet, but with strict safety limits

Anthropic launched Claude Fable 5 on Tuesday, the first Mythos-class model available to anyone. According to the company, it is the most capable AI it has ever made publicly accessible, and it comes with hard limits on what it will answer.

The company said in a statement that in high-risk areas such as cybersecurity, biology, and chemistry, Fable 5 blocks its own responses and falls back to Claude Opus 4.8. Anthropic stress-tested the model with more than 1,000 hours of external red-teaming before release, finding no universal jailbreaks. Some novel attacks may still be possible, the statement added.

Fable 5 is available now through the Claude API and consumption-based Enterprise plans. Through June 22, it is included at no extra cost on Pro, Max, Team, and seat-based Enterprise plans. On June 23, usage credits will be required. Anthropic plans to restore Fable 5 as a standard subscription feature as soon as possible.

Fable 5 and Mythos 5 are priced at $10 per million input tokens and $50 per million output tokens. All traffic is subject to a 30-day retention policy, even for enterprises that previously had zero-retention agreements. Anthropic says the data will not be used for training, only to defend against novel attacks and reduce false positives.

Running alongside this public launch is an expansion of a quieter programme. Project Glasswing was launched in April 2026 as a consortium through which vetted organisations use Mythos to find and fix software vulnerabilities before they can be exploited. The model that powers Fable 5 is the same underlying model as Mythos, made available to the general public for the first time with safety guardrails applied.

Glasswing initially included around 50 partner organisations, among them Amazon, Google, Microsoft, Apple, and Nvidia, alongside cybersecurity firms including CrowdStrike and Palo Alto Networks. The programme pairs those organisations with Mythos to proactively surface critical zero-day vulnerabilities before attackers can exploit them. By late May, early users had identified more than 10,000 serious security vulnerabilities using the model.

On June 2, Anthropic added roughly 150 organisations across more than 15 countries to the programme. India was among them. Indian bodies understood to have received or to be in line for access include CERT-In, the National Critical Information Infrastructure Protection Centre, and the Department of Telecommunications' Digital Intelligence Platform. Major technology firms, including Infosys and TCS, were said to be conducting controlled security tests, while CERT-In was reviewing critical systems including Aadhaar and government login platforms.

The expansion followed weeks of high-level government engagement. Finance Minister Nirmala Sitharaman and IT Minister Ashwini Vaishnaw held a meeting to assess risks posed by Mythos to India's banking sector, resulting in directions to the Indian Banks' Association to develop a coordinated institutional response. SEBI also issued cybersecurity guidelines for regulated entities, including immediate patching, AI-based vulnerability assessments, and zero-trust frameworks. India's inclusion marks a significant step in the internationalisation of what began as a tightly controlled US government programme.

Meanwhile, the race to go public

The Fable 5 launch arrives as Anthropic marches toward its own IPO. The company filed confidentially for a US listing on June 1, days after closing a $65 billion Series H round that valued it at nearly $1 trillion. OpenAI filed confidentially for its own IPO the same week, and AI chipmaker Cerebras recently listed, with SpaceX's offering expected next week. The AI class of 2026 is heading to Wall Street all at once.

Last week, Anthropic President Daniela Amodei, speaking at Bloomberg Tech in San Francisco, said the move to public markets comes down to capital. "It's a really big upfront cost to train the models and to serve inference on them," she said.

Annualised revenue crossed $47 billion in May, up from roughly $9 billion at the end of 2025. Amodei pushed back against sceptics questioning whether AI spending generates adequate returns, saying businesses are still in the early stages of figuring out how to deploy AI effectively. She pointed to coding, financial services, legal services, and healthcare as sectors where value will compound over time.