“the strongest cybersecurity capabilities of any model in the world,” Anthropic wrote on June 9 as it unveiled two new generative-AI models: Claude Mythos 5 and Claude Fable 5.
Claude Mythos 5 and Claude Fable 5 — an overview
Anthropic described Mythos 5 as “an upgrade to Claude Mythos Preview” and positioned the model as a step up in cybersecurity-focused capability. The company released Claude Fable 5 alongside Mythos 5; Anthropic said Fable 5 is powered by the same underlying frontier AI model as Mythos 5 but carries additional guardrails “especially in areas like cybersecurity where the company said it ‘could be misused to cause serious damage.’” The announcement framed Mythos 5 as a high-capability, tightly scoped asset and Fable 5 as a more conservative, broadly accessible variant tuned to reduce potential misuse.
Project Glasswing and initial U.S. government collaboration
Anthropic said Mythos 5 will be deployed initially through Project Glasswing, a cybersecurity-focused frontier AI program the company is expanding. The rollout through Project Glasswing is described as taking place “in collaboration with the US government,” with Anthropic planning a wider availability later via “a broader trusted access program.” That staged approach separates the immediate, restricted use of Mythos 5 from the company’s intention to make similarly capable models available more broadly over time.
Fable 5’s guardrails and Claude Opus 4.8 fallback
To limit risky outputs, Anthropic built a layered safeguard for Fable 5 that, in some cases, routes queries to Claude Opus 4.8 — described as “Anthropic’s next-in-line model that is available to everyone.” Anthropic warned: “To release the model both safely and quickly, we’ve tuned these safeguards conservatively – they’ll sometimes catch harmless requests, though they trigger, on average, in less than 5% of sessions.” The company added it is working on new models where safeguards will be fine‑tuned to reduce false positives. The design effectively creates two tracks: a high-capability Mythos-class model for restricted uses and a guarded, fallback-enabled Fable experience for general availability.
Pricing, distribution, and Microsoft integrations
Anthropic set list prices for Mythos 5 and Fable 5 at $10 per million input tokens and $50 per million output tokens. The company highlighted that those rates are “less than half the price of Claude Mythos Preview,” which it listed at $25 per million input tokens and $125 per million output tokens. Shortly after Anthropic’s announcement, Microsoft said Fable 5 is now available in Microsoft Foundry and “can be used to power agents in GitHub Copilot and Foundry Agent Service,” signaling immediate commercial pathways for the guarded model through major toolchains.
Reactions from cybersecurity leaders: praise, skepticism, and warnings
- Sachin Puri, CEO of Network Solutions, called Mythos-class models “a meaningful step forward in how AI can help identify, analyze and respond to increasingly sophisticated cyber threats and detect vulnerabilities,” while cautioning that deployment must include “governance, accountability and what happens when guardrails fail.”
- Jamie Moles, senior technical manager at ExtraHop, expressed skepticism about public launches of Mythos-class tools “after widespread warnings from safety experts that the model poses severe risks to corporate security,” and urged responsible deployment to avoid compromising systems the public relies on.
- Andrew Rubin, founder and CEO of Illumio, argued that guardrails are not proof the problem is solved: “It’s an admission that even the companies building these models don’t fully trust where the capability leads,” he said, adding that “attackers won’t operate at that layer. They’ll go straight after the capability itself.” Rubin warned that broader availability will increase “the speed and scale of attacks” unless defenders can match that pace.
- Douglas McKee, director of vulnerability intelligence at Rapid7, predicted Fable will become “the everyday workhorse” and Mythos will remain “the power tool for tightly vetted security and infrastructure teams,” urging Anthropic to “keep getting Mythos-class access into the hands of vetted defensive teams through restricted channels and its partner ecosystem.”
What this means for security teams, policymakers, and enterprise buyers
- Security teams: expect a bifurcated toolset — a guarded, broadly available Fable 5 for routine work and a tightly controlled Mythos 5 available through restricted channels like Project Glasswing for high‑risk, high‑reward tasks; teams will have to weigh capabilities against access controls and monitoring.
- Policymakers and government partners: Anthropic’s initial collaboration with the US government and its “broader trusted access program” flag a role for public sector gating and oversight in how Mythos-class capabilities are fielded.
- Enterprise buyers and platform operators: lower list prices for the new models, coupled with Microsoft’s Foundry and GitHub Copilot integrations for Fable 5, create commercial incentives to adopt guarded models rapidly while also raising questions about how effectively those guardrails prevent misuse in practice.
Anthropic’s June 9 release sets a clear tension: it markets Mythos 5 as a leap in cybersecurity capability while steering the wider public toward a guarded Fable 5 that sometimes defers to Claude Opus 4.8. The company says it is tuning safeguards conservatively and plans broader trusted access, but several industry voices warn that interface controls may not be enough if attackers target the capability itself. How Anthropic balances access, pricing, and the protections that security professionals say are necessary will determine whether defenders can keep pace.
