Ethical Hacking News
Cisco praises Anthropic's AI bug hunt model, Claude Mythos Preview, for its transformative power in uncovering cybersecurity breakthroughs. The advanced AI system scanned 1.8 billion lines of code over a period of eight weeks to identify vulnerabilities in Cisco products. With the expanded Project Glasswing program, Anthropic has added 150 new partners, bringing the total number of participating organizations to approximately 200. These advancements signal significant progress in leveraging artificial intelligence for cybersecurity and represent an important step forward in the ongoing quest to protect against emerging threats.
Cisco praises Anthropic's AI bug hunt model, Claude Mythos Preview, for its transformative power in uncovering cybersecurity breakthroughs. The model scanned 1.8 billion lines of code over eight weeks to identify vulnerabilities in Cisco products, with a false positive rate of under 3 percent. The speed of the model is only half the story; its actual contribution lies in the depth and breadth of its discoveries. Human expertise remains essential for validating the accuracy of AI-generated findings, even with a human-guided harness. The model's ability to assess entire code bases without focusing on a specific scope is a significant advancement. This approach provides engineering teams with actionable intelligence rather than just warnings, facilitating the development of effective patches and mitigation strategies. Anthropic expands Project Glasswing, adding 150 new partners, bringing the total to around 200 diverse organizations. New partners report significant cybersecurity findings, including Palo Alto Networks' identification of 26 CVEs representing 75 underlying security issues in a single month.
Cisco has recently praised Anthropic's AI bug hunt model, Claude Mythos Preview, for its transformative power in uncovering cybersecurity breakthroughs. The company, which heads Cisco's security and trust organization, Anthony Grieco, stated that the advanced AI system scanned 1.8 billion lines of code over a period of eight weeks to identify vulnerabilities in Cisco products.
According to Grieco, the "speed is only half the story" when it comes to Mythos' capabilities. He highlighted the "scale, quality, and impact" of the model's findings as the real breakthrough. This indicates that while speed was an important factor in the model's performance, its actual contribution lies in the depth and breadth of its discoveries.
Mythos, which is part of Anthropic's Project Glasswing, was paired with a "human-guided harness" to achieve a false positive rate of under 3 percent. This suggests that while AI can make significant strides in identifying potential vulnerabilities, human expertise remains essential for validating the accuracy of these findings.
The model's ability to assess entire code bases of a product without focusing on a specific scope is a significant advancement. Grieco likened this process to switching from a flashlight to a flood light, illuminating a dark room and providing a clearer understanding of the landscape.
This approach has allowed engineering teams at Cisco to receive actionable intelligence rather than simply being presented with a wall of warnings. The clarity and usefulness of these findings are critical in facilitating the development of effective patches and mitigation strategies for identified vulnerabilities.
Anthropic recently announced an expansion of Project Glasswing, adding 150 new partners to the program. This growth brings the total number of partner organizations under the umbrella to approximately 200. The expanded group includes a diverse range of entities, including government agencies and corporate partners.
Palo Alto Networks, one of the original Project Glasswing partners, reported significant findings after using Anthropic's AI models to scan over 130 products across its three platforms. The company identified 26 CVEs representing 75 underlying security issues within a single month of utilizing the service.
Rubrik, another new partner in the expanded program, stated that it had joined forces with Anthropic to leverage the capabilities of Mythos-class models. This expansion is expected to bring significant benefits to organizations participating in Project Glasswing, including improved cybersecurity and reduced risk.
The newly added partners encompass various industries and countries, with a focus on sectors not well-represented in the initial cohort of Project Glasswing participants. These include power, water, healthcare, communications, and hardware, underscoring Anthropic's commitment to providing comprehensive support for diverse organizations.
Anthropic emphasized that each new partner must meet specific security requirements before gaining access to Mythos. The company has established rigorous standards to ensure that participating organizations adhere to the highest standards of data security and integrity.
Related Information:
https://www.ethicalhackingnews.com/articles/Cisco-Praises-Anthropics-AI-Bug-Hunt-Model-for-Unlocking-Cybersecurity-Breakthroughs-ehn.shtml
https://www.theregister.com/ai-and-ml/2026/06/02/cisco-praises-ai-bug-hunt-wont-reveal-flaw-tally/5250291
Published: Tue Jun 2 14:28:57 2026 by llama3.2 3B Q4_K_M
