Microsoft is reportedly building a cheaper rival to Anthropic's restricted Mythos security model. Does undercutting on price make AI security better or worse?

[ProperJobs89]

Microsoft is said to be preparing Project Perception, an enterprise AI security platform that routes vulnerability hunting tasks across Microsoft, OpenAI and Anthropic models, aiming to deliver similar capability to Anthropic's Claude Mythos at a meaningfully lower cost. Mythos is currently priced at roughly 10 dollars per million input tokens and 50 dollars per million output tokens, with access tightly restricted to a small set of vetted cybersecurity partners due to how capable it reportedly is at finding real exploitable vulnerabilities

The argument for Microsoft's approach is straightforward, advanced AI assisted vulnerability discovery is currently so expensive and so restricted that most organizations simply cannot access it at all, and a cheaper multi model alternative could bring that same defensive capability to a much wider range of businesses who currently have no access to anything like it. The obvious counter argument is that Mythos is restricted precisely because a model this good at finding exploitable weaknesses can just as easily be misused for offense, and a cheaper, more widely available alternative could meaningfully lower the bar for who can access that same capability

This is an open question rather than an obvious answer either way, so worth debating directly, does broadening access to AI powered vulnerability hunting make the internet safer overall because more defenders can patch things faster, or does it make things riskier because the same capability becomes available to a much larger and less vetted pool of users than Anthropic's current gatekeeping allows

News Microsoft Project Perception Ai Security Toolwww.techrepublic.com

[Jonathan11]

Broader access to defensive tooling has to be a net good overall, most of the attackers already have access to plenty of offensive tooling regardless of what Microsoft or Anthropic release

[Glenn_44]

Disagree a bit, the whole reason Mythos is gatekept is presumably because someone at Anthropic ran a real risk assessment and decided narrow access was the safer path, undercutting that with a cheaper alternative skips past that judgment entirely

[TeaAndCode72]

Microsoft using a mix of models including Anthropic's own Claude in the routing is an interesting detail, they are essentially building on top of the very model they are trying to undercut on price

[Cobalt Warren]

Cost has always been the real access barrier for smaller companies wanting serious security tooling, this could genuinely help small and mid sized businesses that Mythos was never going to reach anyway

[TheRock]

The vetting process itself matters more than the price tag honestly, if Microsoft applies similarly rigorous access controls this could be fine, if it is looser to compete on price that is the real risk

[David1]

Feels like classic competitive dynamics, one company gatekeeps a powerful capability, a rival undercuts on price and access, eventually the market decides how much gatekeeping was actually necessary

[ProperJobs50]

Worth remembering Mythos has apparently already found thousands of real critical vulnerabilities in open source projects, so the capability itself is clearly proving useful when properly vetted

[ForumGremlin93]

I lean toward net positive for defenders as long as Microsoft is not being reckless about who gets access, cheaper defensive tooling reaching more companies sounds like a good outcome on balance

[Wardlow]

This whole dynamic of frontier labs racing to build cheaper alternatives to each other's restricted safety tier products is going to become a much bigger story across the whole industry

[Drifter]

The pricing gap alone, 50 percent cheaper according to some reports, tells you Microsoft thinks the restriction was more about caution than genuine unavoidable cost

[Wolfhound39]

Genuinely unresolved question in my mind too, this is exactly the kind of tradeoff that will only really get answered once we see how it plays out in practice over the next year