Clement Delangue wants AI firms to answer for rogue bots after hack

Clement Delangue, the boss of a hacked company, says AI firms must answer for rogue bots. His core complaint is blunt: he does not want cyber attacks on other companies to become “normalised.” In other words, he is arguing that a growing class of AI-enabled threats should not be treated like weather, random and unavoidable, once something starts going wrong.

That framing matters because it is not just about one incident. If rogue bots can be used to probe systems, spread damage, or automate malicious activity at scale, then every company that is not protected in depth ends up paying the price. Delangue’s point pushes the debate away from “who was hacked” and toward “who built the ecosystem that made the hacking easier.” For boards and security leaders, that means the question is no longer only whether an internal team patched fast enough, but whether vendors, platforms, and developers are carrying enough responsibility for how their tools can be abused.

To understand why this is suddenly front and center, look at how AI products are rolling out. Models and automation are increasingly embedded into everyday workflows, which is great for legitimate use. But the same automation that helps customer service also helps an attacker write, iterate, and deploy at speed. The BBC report is specific that Delangue does not want these attacks to blend into business as usual. That “normalised” word is the tell. When an outcome becomes normal, budgets shift, urgency fades, and oversight lags behind capability.

There is also a governance angle. In a lot of enterprises, security incidents trigger internal after-action reviews, incident response costs, and sometimes vendor disputes. But the modern AI question is: what obligations do AI firms have beyond general security claims? Delangue is essentially arguing that accountability should extend to the behavior of tools and agents once they are out in the world. That puts pressure on both product teams and procurement teams. It also creates friction inside organizations, because a vendor may see the threat as misuse, while customers increasingly see it as an expected risk that must be managed upstream.

Regulatory and legal expectations tend to follow lived reality. When an abuse pattern becomes repeatable, regulators often move from broad principles to more concrete requirements, such as auditability, risk management, and clearer accountability chains. Delangue’s stance is consistent with that direction: he is signaling that “we didn’t intend it” may not be enough. The industry will likely be asked to prove not only that models work, but that risk is actively controlled, measured, and prevented from turning into systemic harm for other firms.

For executives, the second-order implications are real. If AI firms do not accept responsibility, the burden flows downstream. That can mean higher insurance premiums, longer sales cycles due to security diligence, and more contractual terms aimed at limiting liability. It can also mean that incident response becomes a core operating capability, not a specialized team function, because the frequency and speed of attacks increases when automation is cheap.

There is another dynamic too: reputational gravity. A hacked company can earn empathy in the short term, but the longer the conversation becomes about “normalised” cyber attacks, the more investors, partners, and customers may wonder what prevention looks like across the stack. Delangue’s message implicitly asks other leaders to compare notes and standards. If rogue bots are emerging as a predictable threat, then boards should treat AI vendor risk as a first-order issue, not an IT checklist item.

The strategic stake for peers is simple. Every company that relies on AI, buys from AI vendors, or builds with AI tools is exposed to the same possibility: that misuse will become routine and defenses will lag. Delangue’s warning is that executives cannot afford to wait until attackers get their process down. If the goal is to keep cyber attacks from becoming normal, responsibility has to move upstream, and accountability has to become part of how AI systems are governed, sold, and monitored.