<span>Story: Technology company Anthropic says its new AI model, Mythos, may be too dangerous to release to the public. </span><span>The unreleased model, designed for defensive cybersecurity tasks, can find and exploit serious software vulnerabilities at a level that approaches, and in some cases surpasses, human experts, the company says.</span><span>This has raised concerns that the same AI capabilities that may aid defenders may also help attackers find new ways to penetrate critical systems.</span><span>:: What are your concerns about Mythos? </span><span>“This particular tool appears to be more powerful than any other tool we’ve seen to date, allowing us to identify vulnerabilities that others, including humans, have missed for decades.”</span><span>Pia Hüsch is a Research Fellow at the Royal United Services Institute. </span><span>“So if this tool is as good as it seems, perhaps attackers could use it to actually launch more powerful attacks than we’ve seen so far, by layering several different layers of vulnerabilities or exploiting multiple sectors in parallel, such as power and water infrastructure.”</span><span>Cybersecurity is a constant competition between those seeking to find and fix weaknesses and those seeking to exploit them. </span><span>Mythos represents a significant step forward in the cyber arms race. </span><span>This could be even more alarming for an already vulnerable industry.</span><span>“I think financial markets are particularly vocal about what this means for the banking sector and its potential to disrupt banking and financial services processes. There are also concerns that this could be used by attackers to exploit vulnerabilities and critical infrastructure, which in many cases already relies on quite outdated infrastructure.”</span><span>:: How has Anthropic mitigated potential risks?</span><span>To limit access, Anthropic has placed Mythos in a restricted effort called Project Glasswing.</span><span>Project Glasswing brings together companies such as Amazon Web Services, Apple, Cisco, CrowdStrike, Google, JPMorgan Chase, Microsoft, NVIDIA, and Palo Alto Networks, as well as more than 40 organizations that build or maintain critical software infrastructure. </span><span>According to Anthropic, the goal of Project Glasswing is to use Mythos to secure critical software before others weaponize similar capabilities. </span><span>“The fact that this has the power to decide which of their chosen partners has access to this really shows how powerful these technology companies are. And right now it’s very unclear who has access to this. And So, I think there are a lot of questions that are left unanswered at the moment, adding to this hype and this perception that it’s an absolute game changer.”</span><span>:: What’s next for cybersecurity? </span><span>The bigger question may not be whether Mythos alone will change cybersecurity, but whether it represents a broader change.</span><span>Anthropic said there is no reason to think Mythos is stagnant and that Frontier capabilities are likely to make significant advances in the coming months. </span><span>CEO Dario Amodei suggested in February that we may be on the cusp of a new era in which computer thinking surpasses human intelligence.</span><span>“We are moving closer and closer to a land of data center geniuses, a set of AI agents that are better at most things than most humans and can adjust at superhuman speed.”</span>
