Anthropic, the AI safety startup founded by former OpenAI researchers, has quietly launched Project Glasswing – an ambitious effort to harden critical software infrastructure against AI-specific vulnerabilities. The initiative comes as governments worldwide scramble to address cybersecurity risks posed by increasingly sophisticated artificial intelligence systems.
According to three sources familiar with the matter, the project involves developing new verification techniques for machine learning pipelines and creating fail-safe mechanisms for core AI infrastructure components. One engineer involved described it as “building circuit breakers for the AI grid” during a recent technical conference.
The timing coincides with heightened regulatory scrutiny. The U.S. National Institute of Standards and Technology (NIST) recently warned that “current software verification methods are inadequate for AI systems” in its 2024 AI Risk Management Framework. Meanwhile, the EU’s AI Act mandates strict cybersecurity requirements for high-risk AI applications.
Market analysts suggest Project Glasswing could position Anthropic as a key player in AI security standardization. “Whoever solves AI infrastructure security will become the plumbing of this technological revolution,” said Maria Vasquez, lead analyst at TechInsight.
However, some experts caution that proprietary solutions may fragment security standards. Dr. Alan Chen of Stanford’s AI Ethics Lab notes: “We’ve seen this movie before with internet security – a dozen competing standards helps no one.”