Anthropic Expands Access to Mythos Through Cisco and IBM Products

The American company Anthropic, developer of the Claude family of AI models, is expanding access to the Mythos model through Cisco and IBM products. The companies announced on November 9–10, 2026, plans to integrate the technology into corporate cybersecurity systems. This will allow organizations to use Mythos's capabilities to find software vulnerabilities without direct access to the model itself.

Previously, Anthropic provided limited access to Mythos through the Project Glasswing program. Participants could use the model to check software code for vulnerabilities. The company continues to control access due to the risk of the model's capabilities being used for developing and conducting cyberattacks.

Cisco to Add Mythos to AI Defense System

The American technology company Cisco announced the integration of Mythos into its AI Defense platform at a conference in Melbourne on November 10. The system is designed to identify vulnerabilities and potentially dangerous behavior of models, AI agents, and applications.

AI Defense already uses artificial intelligence to simulate attacks on systems under review. Mythos will complement existing tools and be used for developing testing scenarios.

Jeff Schultz, Cisco's Senior Vice President of Product Portfolio Strategy and Marketing, stated that the integration will allow the system to identify additional threats.

However, clients will not be able to independently send arbitrary requests directly to Mythos. Testing will be limited to systems for which the organization has permission to review. The actions of the tools will be logged, and tests can be halted.

According to information obtained by journalists from The Register, Cisco intends to provide the new functionality within existing licenses at no additional cost. However, detailed public terms have not yet been presented.

Initially, the integration is planned to be opened to select corporate clients. The company has not announced dates for a broader launch.

IBM Plans to Implement Mythos in Concert Protect

IBM, an American technology company, also announced its intention to use Mythos in its Concert Protect platform, designed for managing vulnerabilities in corporate applications.

The model will analyze software code, identify potential vulnerabilities, and help determine which ones could actually be exploited by attackers. The system will also be able to suggest fixes, which developers must then approve.

IBM plans to make the new functionality publicly available in early 2027. Exact terms of use have not yet been disclosed.

Anthropic Updates Mythos and Fable to Version 5.2

On November 9, Anthropic introduced Claude Mythos 5.2 and Fable 5.2. Both versions use the same base model but differ in protective restrictions and access conditions. Fable is available to a wide audience, while using Mythos for specialized tasks requires organizational verification.

According to Anthropic, the updated models generate responses up to 30% faster than previous versions. The cost is $8 per 1 million input tokens and $40 per 1 million output tokens.

An additional fast mode allows increasing generation speed up to 2.5 times. When using it, the cost rises to $16 and $80, respectively.

Anthropic also reported conducting additional security tests, including checking the possibility of programs escaping the isolated execution environment. According to the company, the risk of serious consequences due to the model's behavior not meeting specified requirements remains low.

The developer claims that in Fable 5.2, restrictions on potentially dangerous tasks in cybersecurity and biology are maintained, but the number of erroneous blocks has been reduced. Protection against attempts to copy the model's capabilities has also been strengthened.

The integrations with Cisco and IBM are still in the implementation stage. Specific results of their application in corporate systems and timelines for expanded access have not yet been presented.