Anthropic Joins Rivals to Prevent AI Cyberattacks
▼ Summary
– Anthropic formally announced its new Claude Mythos Preview AI model and a related industry consortium called Project Glasswing to address cybersecurity implications.
– The Project Glasswing consortium includes major tech firms like Microsoft, Apple, and Google, giving them private access to the model to find and fix vulnerabilities in their own systems.
– Anthropic states the model’s advanced coding ability inadvertently makes it proficient at cybersecurity tasks like finding vulnerabilities and developing attack chains.
– The company is using a staggered release, starting with this consortium, to mirror coordinated vulnerability disclosure and allow for defensive preparations before broader availability.
– The initiative aims to prepare for a near-future where such broadly available AI capabilities could fundamentally disrupt current global security practices and assumptions.
In a significant move for AI cybersecurity, Anthropic has initiated a private industry consortium alongside the announcement of its new Claude Mythos Preview model. This collaborative effort, named Project Glasswing, brings together major technology firms, cybersecurity experts, and critical infrastructure organizations to proactively address the security challenges posed by rapidly advancing artificial intelligence. The initiative recognizes that powerful new AI capabilities demand a coordinated response to safeguard global digital systems before these tools become widely accessible.
The consortium includes Microsoft, Apple, Google, Amazon Web Services, and dozens of other entities from the tech, finance, and infrastructure sectors. These partners will receive exclusive early access to the Mythos Preview model. A core objective is to provide these foundational platform developers a crucial window to test the AI against their own systems. This allows them to identify and remediate potential vulnerabilities and exploit chains that the model might uncover through simulated attacks, essentially stress-testing digital defenses in a controlled environment.
Anthropic’s leadership stresses that the project’s urgency extends far beyond their own model. They aim to spark industry-wide preparation for a near-future where advanced AI fundamentally alters the security landscape. “The real message is that this is not about the model or Anthropic,” said Logan Graham, the company’s frontier red team lead. “We need to prepare now for a world where these capabilities are broadly available in 6, 12, 24 months. Many of the assumptions that we’ve built the modern security paradigms on might break.”
This development highlights a new phase in the perpetual cat-and-mouse game of cybersecurity. AI models from various companies are now proficient at analyzing code to find weaknesses, suggest fixes, or even devise methods for exploitation. While such tools can empower defenders, they equally risk lowering the barrier for malicious actors, making sophisticated attacks more feasible and scalable.
Anthropic CEO Dario Amodei emphasized the unexpected potency of the new model in a Project Glasswing launch video. “Claude Mythos preview is a particularly big jump,” he stated. “We haven’t trained it specifically to be good at cyber. We trained it to be good at code, but as a side effect of being good at code, it’s also good at cyber.” He added that more powerful models are inevitable, necessitating a strategic plan from the entire industry.
According to Graham, Mythos Preview demonstrates capabilities that rival a seasoned security researcher. Its skills extend beyond simple vulnerability discovery to include advanced exploit development, penetration testing, and binary analysis. This level of proficiency directly informs Anthropic’s cautious release strategy. The company is modeling its approach on coordinated vulnerability disclosure, granting trusted partners time to develop patches and defenses before a broader rollout. “Done not carefully, this could be a meaningfully accelerant for attackers,” Graham warned.
Participants in Project Glasswing, including direct competitors of Anthropic, have endorsed the collaborative spirit of the initiative. In a supporting statement, Google’s Vice President of Security Engineering, Heather Adkins, noted, “Google is pleased to see this cross-industry cybersecurity initiative coming together. We have long believed that AI poses new challenges and opens new opportunities in cyber defense.” This unified stance underscores the shared recognition that the security implications of advanced AI transcend individual corporate interests.
(Source: Wired)
