Venture capitalist Simon Dedic says Anthropic’s latest AI model lowers the barrier to crypto exploitation to “basically zero.” He made the claim in response to the release of Anthropic’s Claude Mythos, a set of new models positioned with safeguards for misuse.
The tension for crypto users is straightforward. Safeguards aim to reduce harm from the model’s output. But if the model also reduces the cost and skill needed to find exploits, it can accelerate the pace of discovery. That cuts two ways. It can help defenders harden systems. It can also help attackers iterate faster, especially when they already know where to look.
Dedic’s specific point is about capability access. He frames the change as economic, not just technical. Lower cost means more attempts. More attempts means more coverage. In security work, coverage often beats cleverness.
What Dedic claims changed
Dedic told Cointelegraph that Anthropic’s model drops the cost and skill required to find crypto exploits to “basically zero.” He is not describing a lab-grade breakthrough for one niche vulnerability. He’s describing a shift in the effort curve.
That matters because crypto exploits often share a pattern. Attackers start with a weak assumption. They test variations until one hits. If the testing harness becomes cheaper, the number of variations tested can spike.
Safeguards versus output reality
Anthropic’s “safeguards” are meant to limit misuse. But the security problem here is downstream. Even restricted systems can produce insights that still help a determined person map risk.
For crypto users, the practical question is not whether the model can be used “in theory.” It’s whether the safeguards reduce the odds that attackers learn faster than defenders. The source text does not provide details on what safeguards Anthropic uses, how they perform, or whether they address exploit-finding workflows specifically.
That gap is important. Without specifics, the claim that safeguards reduce exploitation risk can’t be verified from the information provided.
The unanswered questions security teams will ask
Based on Dedic’s claim, crypto-focused teams will likely look for evidence on three fronts. How quickly exploit discovery could change if attackers use AI-assisted search. Whether safeguards block exploit development steps or only the most obvious requests. Whether Anthropic measures jailbreak success and how it handles adversarial prompting aimed at technical vulnerability patterns.
Cointelegraph’s provided excerpt stops at Dedic’s “basically zero” framing and does not include Anthropic’s technical safeguards description. So readers are left with a warning signal and not enough data to quantify it.
Why the timeline matters
In security, the first mover advantage often belongs to whoever can iterate fastest. Dedic’s economic framing suggests iteration could get cheaper. That alone doesn’t guarantee more attacks. But it increases the number of opportunities attackers can test.
Defenders will respond by tightening review and expanding monitoring, especially around attack surfaces where small logic errors become big losses. The risk is not that every exploit gets found and used. The risk is that the time window between “known risk” and “weaponized exploit” shrinks.
Cointelegraph reports Dedic’s claim and pairs it with the context of safeguards around Claude Mythos. The next step for anyone responsible for crypto security is to demand concrete details. Safeguards need measurable performance. If they can’t be measured, they can’t be relied on.