Skip to content
Inside the Cyber Defence of Claude Fable 5.2 and Mythos 5.2

Inside the Cyber Defence of Claude Fable 5.2 and Mythos 5.2

Cybermagazine • November 10, 2026

Anthropic has unveiled two new AI models, Claude Fable 5.2 and Claude Mythos 5.2, to tackle complex technical challenges across industry workflows.

The launch follows the rollout of Claude Haiku 5.5 in October as the business broadens its product portfolio ahead of an intended IPO.

Anthropic just released Mythos 5.2 and its publicly available sibling Fable 5.2. Same intelligence, different safety guardrails. 30% faster and 20% cheaper than the gen. Seven months ago this class of models was considered too risky for public release Oleg Zaremba, Co-Founder and CTO at AiSDR

Anthropic just released Mythos 5.2 and its publicly available sibling Fable 5.2. Same intelligence, different safety guardrails. 30% faster and 20% cheaper than the gen. Seven months ago this class of models was considered too risky for public release

Competition across the sector continues to grow as OpenAI, Alphabet’s Google and Meta Platforms introduce frontier systems.

These competing firms are releasing tools designed for coding, reasoning and autonomous agents to attract developers and business clients.

In a post made on , Oleg Zaremba , Co-Founder and CTO at AiSDR , says: “Anthropic just released Mythos 5.2 and its publicly available sibling Fable 5.2. Same intelligence, different safety guardrails. 30% faster and 20% cheaper than the gen. Seven months ago this class of models was considered too risky for public release.”

Cybersecurity engineering advances through verification programmes

Claude Mythos 5.2 targets defensive cyber engineering alongside approved life-sciences research.

Access to Mythos remains restricted to verified organisations through Anthropic’s Cyber Verification Program and Life Sciences Verification Program.

Claude Enterprise users can access the technology through Claude Security to inspect software repositories.

The security tool tracks code adjustments, learns contextual data and links to development workflows through an API and MCP.

In CVE Rediscovery assessments, Claude Mythos 5.2 identified 10.2% more vulnerabilities than Claude Opus 5.5. The model detected 14.4% more software flaws than Claude Mythos 5.1 during the same trials.

External evaluations by a security firm revealed that one in six flaws confirmed by Mythos 5.2 were missed by competing models.

Evaluators also reported that the release eliminated overstated bug severity, which previously affected one in 20 findings.

30% faster output: Both systems generate outputs up to 30% faster than their 5.1 predecessors

20% lower cost: The models run at a 20% lower cost than the generation

10.2% more vulnerabilities: Mythos 5.2 detected 10.2% more vulnerabilities than Claude Opus 5.5 in CVE Rediscovery assessments

14.4% more flaws: Mythos 5.2 uncovered 14.4% more software flaws than Claude Mythos 5.1 in internal testing

One in six flaws: An external security company found that one in six confirmed vulnerabilities discovered by Mythos 5.2 were missed by all other tested frontier models

Zero overstated findings: The model eliminated exaggerated severity ratings, down from one in 20 findings (5%) in prior editions

Model safeguards support varied deployment roles

Claude Fable 5.2 and Claude Mythos 5.2 one underlying build designed for reasoning, vision and coding tasks.

Fable incorporates biology and cyber safeguards with fewer unnecessary blocks, allowing wider deployment without misuse.

During trials, the shared architecture resolved two physics problems, formalised mathematical proofs and designed three operational electronic devices.

Organisations can deploy Fable as an Advisor to assist other models or as an Orchestrator delegating smaller tasks. Outputs from the systems process up to 30% faster than 5.1 editions.

Both systems cost US$8 per 1m input token and US$40 per 1m output tokens. An optional fast mode increases token generation speed by up to 2.5 times.

The accelerated mode costs US$16 per 1m input tokens and US$80 per 1m output tokens.

Government safety bodies have also evaluated the new releases to assess how frontier models behave under oversight controls.

Kanishka Narayan MP , who is the Minister for Artificial Intelligence in the UK , notes that pre-deployment assessments revealed Claude Mythos 5.2 proved significantly more capable of evading supervisory monitoring systems than earlier models.

Kanishka notes: “AISI tested Mythos 5.2 for its monitorability: assessing how effectively models can be caught by monitors, which can oversee and block actions. Monitorability is an important safety property.

“AISI found that Mythos 5.2 was significantly more capable of evading certain types of monitors used in deployment than Anthropic models tested. Developments like these mean that models may be getting harder to monitor and AISI is closely following this trend.”

He adds: “AISI will continue to work closely with industry partners, including Anthropic and research like this is precisely how we drive forward the critical work of making AI models safer.

“AISI’s pre-deployment testing keeps the UK at the forefront of understanding frontier AI capabilities. Rigorous testing like this is the path to safer AI that we can all benefit from.”

Key strategic partners

Amazon : A multinational technology enterprise that serves as Anthropic’s primary cloud partner through Amazon Web Services. The company provides custom Trainium silicon alongside scalable infrastructure for frontier model training and distributes Claude systems to enterprise customers globally via Amazon Bedrock, helping organisations deploy secure generative AI applications across complex technical workflows.

Google : A global technology company that supplies Anthropic with substantial cloud computing capacity and specialised Tensor Processing Unit infrastructure for AI workloads. The organisation has invested significant capital in the developer while offering Claude models to corporate clients via Google Cloud’s Vertex AI platform to accelerate enterprise automation and secure digital operations.

Infosys : An international digital services and consulting corporation that collaborates with Anthropic to develop tailored artificial intelligence agents for enterprise clients. Operating a dedicated Centre of Excellence, the organisation deploys specialised Claude workflows across telecommunications and other regulated sectors to assist businesses with technical software migrations, customer operations and secure digital transformation.

Kanishka Narayan UK AI Minister

Oleg Zaremba Co-Founder and CTO

AI Misalignment: How Anthropic's AI Hacked a Fourth Company Technology & AI

Inside Anthropic's Claims of Distillation Attack by Alibaba Technology & AI

Why the US is Freezing Anthropic’s New Claude Models Cyber Security

Fujitsu's Dual AI Deal: Claude for Defence & ChatGPT for Ops Technology & AI

Extracted Entities

Companies (1)

Industries (1)

Tools (1)