Key Points
- Zscaler joins Anthropic's Project Glasswing and OpenAI's Trusted Access for Cyber programme
- US Treasury Secretary met Anthropic CEO to discuss AI security implications
- RBI considering engagement with regulators on AI model risks to financial systems
Zscaler has joined two major artificial intelligence cybersecurity initiatives run by Anthropic and OpenAI, as the cloud security firm seeks to counter threats from AI systems capable of detecting and exploiting software vulnerabilities faster than human security teams.
The San Jose-headquartered company said it is participating in Anthropic’s Project Glasswing and OpenAI’s Trusted Access for Cyber (TAC) programme. Both initiatives aim to combine advanced AI models with enterprise security systems to identify software flaws and strengthen cyber defences before attackers can exploit them.
Frontier AI models — the most advanced AI systems developed by leading laboratories — can now chain together multiple software vulnerabilities and generate working exploits at speeds that outpace traditional security testing conducted by human analysts.
How Zscaler will deploy the AI models
Under Project Glasswing, Zscaler will integrate Anthropic’s Claude Mythos Preview model into its software development process to detect vulnerabilities earlier. The company said the integration would strengthen its Zero Trust Exchange platform, a security architecture that assumes no user or device should be trusted by default and requires continuous verification before granting access to applications or data.
Under OpenAI’s TAC programme, Zscaler will deploy GPT-5.4-Cyber across its internal security systems and threat detection workflows. The TAC programme gives vetted cybersecurity organisations controlled access to advanced AI models, allowing them to build defensive tools while limiting the risk that the same models could be misused by malicious actors.
Why Project Glasswing was launched
Anthropic conceived Project Glasswing after AI systems demonstrated the ability to autonomously identify and exploit software vulnerabilities at scale. The capability raised concerns that malicious actors could use the same technology to attack organisations faster than defenders could respond.
The initiative brings together AI laboratories, cybersecurity firms and enterprise technology companies to test these models in controlled environments, develop safeguards and ensure defensive capabilities evolve alongside offensive risks.
Early participants in such collaborations have included Palo Alto Networks, CrowdStrike, Cloudflare and Microsoft, working with AI developers to evaluate how frontier models behave in real-world attack scenarios.
US government engagement with AI firms
The expansion of these programmes comes as the US government deepens engagement with AI firms on security risks. US Treasury Secretary Scott Bessent and White House Chief of Staff Susie Wiles recently met Anthropic CEO Dario Amodei to discuss the implications of advanced AI systems, including cybersecurity and broader national security concerns.
The engagement is notable given ongoing policy debates in the US around limits on military applications of AI. The Department of War has banned Anthropic from certain defence contracts, yet federal agencies are working with AI developers to understand risks emerging from frontier models.
The meetings signal closer coordination between governments and AI companies as capabilities evolve and potential misuse scenarios expand.
RBI examining AI risks to financial systems
Regulators globally are examining the role of AI in cybersecurity frameworks, particularly for protecting critical infrastructure and financial systems. In India, the Reserve Bank of India is considering engagement with other financial regulators and global counterparts to understand the implications of advanced AI models on financial system stability and fraud risks, according to a Reuters report.
The central bank’s interest showa concern that AI-driven tools could both strengthen fraud detection and be misused for sophisticated financial attacks, requiring coordinated policy responses across jurisdictions.
Zscaler said combining frontier AI models with its Zero Trust architecture would reduce exposure to cyber threats by eliminating internet-facing attack surfaces and strengthening application security. The company added it will continue integrating AI models into its red teaming, threat detection and vulnerability management systems.
Your Questions, Answered
What is Project Glasswing?
Project Glasswing is an Anthropic initiative that brings together AI laboratories, cybersecurity firms and enterprise companies to test advanced AI models in controlled environments and develop safeguards against AI-enabled cyber threats.
What AI models will Zscaler deploy?
Zscaler will integrate Anthropic's Claude Mythos Preview model under Project Glasswing and deploy OpenAI's GPT-5.4-Cyber under the Trusted Access for Cyber programme for threat detection and vulnerability management.
Why are frontier AI models a cybersecurity concern?
Frontier AI models can autonomously detect software vulnerabilities, chain multiple flaws together and generate working exploits faster than human security teams, raising concerns about misuse by malicious actors.
Is the RBI examining AI cybersecurity risks?
According to a Reuters report, the Reserve Bank of India is considering engagement with financial regulators and global counterparts to understand how advanced AI models could affect financial system stability and fraud risks.
