Buterin concedes other labs will build top-end offensive cyber tools anyway, so the filter is meant to buy time, not stop them for good.
Ethereum co-founder Vitalik Buterin said in a post on X on Monday that he is comfortable with the AI model Kimi 3.1 deliberately leaving top-end offensive cyber capabilities out of its training data.
In his view, the trade-off is worth making because it gives defenders more time to prepare.
Restricting Use Cases, Not Users
“Discriminating against use cases is MUCH less bad than discriminating against users,” Buterin wrote.
The world, he added, needs time to harden systems and check old code before “the full weight of very powerful cyber attacks is upon us.” In plain terms, he prefers a model that declines certain tasks for everyone over one that turns certain people away.
He also sketched what could go wrong. If “this first big wave of AI risk is not handled well,” he cautioned, the result might be a “dystopia of highly locked down models,” with KYC checks, which tie an account to a verified identity, required for everything and intrusive surveillance to match. Depending on credentials and nationality, a person could end up a “second or third or fourth-class citizen” in terms of model access.
Asked by X user LOID whether other labs would build those capabilities anyway, Buterin replied: “Yes of course they will.” Read alongside that argument about time, the filtering is meant to delay such capabilities rather than prevent them.
Not everyone agreed. “Hard disagree,” wrote developer Youssof Al Toukhi, who brought up the recent attack on Hugging Face by an OpenAI model, and when Hugging Face tried to analyze the attackers’ actions using frontier models, the providers’ safety guardrails blocked the requests, forcing a switch to an open model, GLM 5.2. Calling the arrangement an “elitism system” that “will only get worse overtime,” the critic added: “Intelligence needs to be democratized.”
Earlier Remarks on AI and Code Security
As CryptoPotato reported previously, Buterin countered fears in May that AI-powered bug hunting would overwhelm developers and create non-stop openings for exploits, arguing that AI-assisted formal verification could become one of the strongest defenses in crypto.
You may also like:
In his telling, a model can now write both the code and its proof, leaving developers to confirm that the statement proven is the one they wanted. His Monday post names formal verification of new codes among the work that needs the extra time. That May 19 post appeared as a hacker took more than $76 million from the Echo Protocol cross-chain bridge, the third major exploit in four days.
In October 8, he warned that AI-assisted mathematics could weaken ECDSA and similar cryptographic systems sooner than expected, although he added that users should not “scramble to move their funds to new wallets today.”