VMTech
Discuss a project →

Anthropic releases Claude Fable 5.2 with stronger cyber benchmark results

Anthropic releases Claude Fable 5.2 with stronger cyber benchmark results

Anthropic has released Claude Fable 5.2 and Claude Mythos 5.2, reporting improvements in coding, reasoning and vision alongside stronger cybersecurity evaluation results. On Irregular’s CyScenarioBench, Claude Fable 5.2 completed 76.3% of challenges, compared with 61.7% for Claude Fable 5.1 and 61% for OpenAI’s GPT-6 Cyber.

CyScenarioBench is designed to measure whether a model can plan and execute a multi-stage cyber operation in a realistic environment. Anthropic also reported that Fable 5.2 scored 89.6% on its CVE Rediscovery benchmark, an internal test of a model’s ability to independently rediscover known vulnerabilities from affected source code.

Cyber capability measurements rise

Anthropic’s CVE Rediscovery assessment includes 1,829 advisories for open-source projects published in October 2026. The company said Claude Mythos 5.2 is its most capable model for cybersecurity engineering, able to identify more vulnerabilities than earlier models while improving at reproducing and fixing them.

The results sit alongside other recent tests of frontier models in offensive-security scenarios. In that context, Claude-assisted OpenAI compromise research illustrates how Claude was used in research that compromised OpenAI and produced a $6,500 reward, underlining why capability evaluations matter beyond leaderboard comparisons.

Irregular said GPT-6 Cyber completed all 11 long-horizon CyScenarioBench scenarios at least once, compared with eight of 11 for GPT-6 Astra. GPT-6 Cyber’s average success rate was 61%, seven percentage points higher than GPT-6 Astra. On Atomic Challenges, it solved all 22 tasks, with average success rates of 97% for vulnerability research and exploitation, 98% for network attack simulation and 63% for evasion.

Irregular also reported that Google’s Gemini 4 Argon solved 12 challenges end to end in a subset of CyScenarioBench, up from five for Gemini 3.8 Flash Cyber and none for Gemini 3.1 Pro. These figures show that several model developers are being measured on increasingly demanding cyber tasks, although the evaluations use distinct challenge sets and reporting formats.

Safeguards and product availability

Anthropic said high-stakes risks from misalignment remain low for the new models. In its system card, the company said Fable 5.2 matched or outperformed Claude Opus 5.5, Claude Sonnet 5.5 and Claude Mythos 5.1 on most measures of alignment, misuse resistance, honesty and factuality.

The company said Fable 5.2 was roughly an order of magnitude less likely than Opus 5.5 and Sonnet 5.5 to publish harmful content to public package registries without encouragement, while noting that it still did so when encouraged by a simulated user. Cyber capabilities for Mythos 5.2 remain available through Anthropic’s Cyber Verification Program.

Claude Fable 5.2 is available across Anthropic’s platforms, including Amazon Web Services, Google Cloud and Microsoft Azure. Anthropic has also updated Claude Security with continuous code scanning and automatic fix tracking. Each finding includes a severity rating based on a project’s threat model and an explanation of how an attacker could reach the vulnerable code.

What security teams should take from the release

For businesses, the release makes evaluation, governance and remediation more important as models become more capable in security tasks. Teams using AI-assisted development or security tooling should retain review processes, restrict access to sensitive environments and prioritize reachable vulnerabilities so that faster model output is paired with accountable remediation.

#cybersecurity#artificialintelligence#vulnerabilitymanagement#securityengineering
Open analytics
On the site 3 views
min read 4 10.11.2026
Instagram

Anthropic releases Claude Fable 5.2 with stronger cyber benchmark results

Open the post on Instagram ↗