AI Sec Watch

Anthropic released Claude Fable 5, a powerful AI model based on its restricted Mythos architecture, with built-in safeguards to make it safely available to the general public. The safeguards work by automatically routing requests about cybersecurity, biology, chemistry, and other high-risk topics to a less capable model (Claude Opus 4.8), though early testing suggests these safeguards may be broader than intended and sometimes block benign requests. Anthropic developed AI-powered classifiers (systems that categorize requests) to identify and block potentially dangerous requests, and says internal and external testing found no effective jailbreaks (methods to bypass security restrictions) that could consistently get around these protections.

ARM announced CVE-2025-10263, an architectural vulnerability in some ARM processor cores that allows attackers to bypass translation stages (memory protection mechanisms that control which parts of memory different software can access) or GPT protections under certain conditions. An attacker running at a lower privilege level can write to memory that should only be accessible to higher privilege software, allowing them to escalate their access rights, though reading protected memory is not affected by this bug.

This academic paper presents a new method for detecting unusual network activity using parallel GANs (generative adversarial networks, AI systems that learn patterns by comparing real data against artificially generated data) without requiring manually set detection thresholds (cutoff points that decide what counts as suspicious). The approach uses comparative reconstruction error learning, meaning it compares how well the AI can recreate normal network behavior to spot deviations that might indicate attacks or intrusions.

Anthropic released Fable, a version of its AI tool that the company previously said was too powerful for public use, though it included safeguards and user limitations. The company also gave access to Claude Mythos 5 (a more capable version without certain restrictions on cybersecurity or biology topics) to a small group of cyberdefenders and infrastructure providers, with plans to expand access further soon.

AI systems are now used in everyday work, and investigators need structured ways to understand what happened when problems occur. Microsoft has published a playbook that helps security teams investigate activity in Microsoft 365 Copilot and Azure AI services (cloud-based AI tools) by using telemetry (data about system activity) collected across Microsoft security products. The playbook uses a scope-context-signal approach: first identifying who used the AI system and when, then checking what data was accessed, and finally evaluating suspicious signals like prompt injection attempts (tricking AI by hiding instructions in its input) or unusual usage patterns.

CVE-2026-45482 is a path traversal vulnerability (a flaw where an attacker can access files outside the intended directory by manipulating file paths) in GitHub Copilot and Visual Studio Code that allows an unauthorized attacker to bypass a local security feature. The vulnerability has a CVSS 4.0 severity score (a 0-10 rating of how severe a vulnerability is, where higher numbers mean more serious). Details are still being assessed by NIST, and Microsoft has published information about this issue.

Anthropic released Claude Fable 5, a powerful AI model similar to its earlier Mythos model, to the public after initially limiting access due to safety concerns. The company implemented new safeguards (filters that block responses in high-risk areas like cybersecurity and biology) to allow the broader release while maintaining security, and also launched Claude Mythos 5, which is the same underlying model but with some safety restrictions removed.

Anthropic released Claude Fable 5, a powerful AI model with safety restrictions that automatically switch to a less capable version when users try to use it for high-risk tasks like cybersecurity or biology. The company tested these safeguards extensively through internal testing and external bug bounty programs (paying security researchers to find vulnerabilities) spanning over 1,000 hours, and no universal jailbreaks (methods to bypass the restrictions) were discovered.

Anthropic released two new AI models: Claude Mythos 5 (limited to industry partners and government collaborators) and Claude Fable 5 (publicly available). Because Mythos 5 can design hacking tools to find software vulnerabilities, Claude Fable 5 includes guardrails (safety restrictions built into the system) that block questions about cybersecurity, biology, and chemistry by routing them to an older, less capable model instead, while Anthropic works on more precise safeguards for future releases.