Advisories
Anthropic Expands Tiered Cyber LLM Access, Limiting Guardrails for Vetted Defenders
Anthropic has merged Project Glasswing into its Cyber Verification Program, extending access to advanced LLMs—Opus, Sonnet, and Mythos—to a vetted group of over 40 organizations while adjusting guardrails. Defenders should verify program enrollment, review guardrail settings, and monitor for misuse.
What Happened
Anthropic merged its Project Glasswing into a tiered Cyber Verification Program (CVP), expanding access to advanced cyber LLMs—Opus, Sonnet, and the newly added Mythos—while adjusting guardrails for vetted security professionals.
Impact
- Access now limited to a vetted group of >40 organizations, including Amazon, Apple, Microsoft, Google, and others.
- Mythos can accelerate vulnerability discovery but carries dual‑use risk; guardrails may be reduced for higher tiers.
- Defenders outside the program will not receive the same level of AI assistance.
What to Do
- Verify if your organization is enrolled in CVP and determine your tier level.
- Review the guardrail settings for the LLM you plan to use and ensure they align with your internal security policies.
- Implement monitoring for anomalous AI‑generated activity that could indicate misuse.
- Keep abreast of Anthropic’s policy updates and adjust access controls accordingly.
Mitigation Guidance
Defenders should treat AI‑driven vulnerability discovery as a high‑value asset but also a potential vector for malicious exploitation. Apply least‑privilege access, audit logs, and cross‑team reviews before deploying Mythos or other advanced models.
Related articles
Anthropic Incident Documentation Highlights LLM CAPTCHA Processing Failures
Recent incident logs from Anthropic reveal that advanced Claude models still struggle to reliably solve standard image-based CAPTCHAs. Defenders should understand that while AI capability scaling continues, automated abuse vectors involving visual verification challenges remain constrained by logical loops and context-handling errors.
Read briefing
CISA & Australia Issue Joint Guidance on Isolating OT and Enabling Systems in Critical Infrastructure
The Cybersecurity and Infrastructure Security Agency (CISA) has teamed up with Australian authorities and other partners to publish a comprehensive set of best‑practice recommendations for segregating operational technology (OT) and enabling systems from enterprise IT networks. The guidance stresses the importance of network segmentation, strict remote‑access controls, continuous monitoring, and incident‑response readiness to protect critical infrastructure from cyber threats that could arise from interconnected systems.
Read briefing
LLM Safety Classifier Evasion Exploited: What SecOps Must Know
CrowdStrike’s Cyber Superintelligence Lab discovered that the leading publicly available content safety classifiers can be bypassed through a request‑aggregate‑bypass technique. The attack allows malicious actors to inject disallowed content into LLM outputs, undermining automated moderation. Defenders should augment LLM pipelines with layered filtering, monitor for anomalous request patterns, and enforce strict content‑policy enforcement.
Read briefing
LLM Safety Classifier Bypass Revealed: Attackers Circumvent CrowdStrike’s AI Filters
The CrowdStrike Cyber Superintelligence Lab exposed a systematic bypass of the most advanced publicly available content safety classifier. Attackers can use request manipulation and aggregation to slip disallowed content past AI filters, potentially undermining automated threat detection and content moderation. Defenders should layer multiple models, enforce input validation, monitor for anomalous patterns, and update classifiers with adversarial data.
Read briefing