AI security experts warn that current defenses are losing to attackers—Mythos raises stakes with new vulnerability research

Security leaders gathered in Washington to debate how to defend AI systems in an environment they acknowledge 'favors attackers'—meaning it's currently easier to break AI than to secure it. The discussion underscores that no consensus exists yet on which defenses actually work against the evolving threats.

Mythos (an AI security company) escalated the conversation by publishing new research on AI vulnerabilities, adding concrete evidence to abstract concerns about where attacks are succeeding. This shifts the debate from 'AI security is important' to 'here are the specific ways AI systems fail today.'

For anyone building or deploying AI tools—startups, enterprises, or teams rolling out chatbots or automation—this means your current security practices are likely incomplete. You now have a clearer map of what to defend against, but also clearer proof that industry-standard protections have gaps.