Recent high-profile security incidents, starting with Hugging Face in August 2026, have raised an uncomfortable question for security leaders. Namely, when an AI agent can move across systems at machine speed, exploiting weaknesses and crossing organizational boundaries, who is actually accountable for preventing, containing and responding to the incident?
That question has become harder to answer as scrutiny of security leaders has increased. Recent high-profile CISO liability cases have put personal accountability firmly on the agenda, while incidents increasingly span systems and teams that do not necessarily sit within the CISO’s direct control.
CISOs must be held accountable for the security decisions, programs and risks within their remit. Yet, the vulnerability might sit within one team, the affected infrastructure within another, and the tactical response with a third. Accountability therefore, cannot mean making one individual responsible for every decision that contributes to an incident.
Yet that is increasingly the tension organizations are facing. Fastly’s 2026 Global Security Research Report found that 73% of organizations say the CISO is ultimately accountable when a breach occurs, while responsibility for managing incidents is spread across the business, most commonly among IT operations and cybersecurity teams.
That creates a clear accountability gap, and many organizations know it. Just 43% say ownership is clear, while exactly the same proportion say it is not.
That leaves organizations evenly split on whether responsibility is clear at all, suggesting many still lack a shared understanding of who owns the response when a breach happens.
Accountability is Centralized, Execution is Not
A CISO might lead the response, but it's rare they execute every part of the response themselves. Fastly’s research shows how widely that responsibility is shared, with IT operations involved in incident management at 69% of organizations and cybersecurity teams at 64%.
In many cases, both teams will play a role, making clear lines of ownership all the more important. Depending on the nature of incidents themselves, the actual tactical responses can sit elsewhere again. For a DDoS or account takeover attack, mitigation may be handled by engineering rather than information security teams.
The problem arises when responsibility for outcomes is centralized around one executive while the people, budgets, systems and decisions needed to deliver those outcomes are distributed elsewhere.
The Rise of Defensive Security Governance
Organizations are responding to increased scrutiny, with 94% making policy changes around CISO liability. Many of those changes are constructive, including greater board involvement, clearer role definition and more support for security leaders. Others focus more heavily on legal protection and disclosure processes when an incident occurs.
Both have a place. The risk comes when defensive measures start to outweigh changes that actually improve security. Stronger legal support and clearer disclosure processes can help manage accountability, but they do not patch systems, change engineering decisions or give security teams the resources to address known risks.
That matters when CISOs are carrying greater personal accountability while responsibility for incident response remains spread across the business. Recent discussion around CISO burnout shows the pressure this can create when governance has not kept pace with expectations.
Boards must therefore look beyond whether a CISO identified a risk and ask whether the organization gave them the means to address it, and who ultimately accepted that risk if action was deferred.
AI is Exposing the Governance Gap
As models become agents, they are connected to more of an organisation’s infrastructure, including APIs, code repositories, cloud environments, data stores and internal tools. Each connection increases the potential attack surface and the blast radius when controls fail.
The UK’s National Cyber Security Centre (NCSC) has warned that integration of AI across technology infrastructure is increasing the attack surface, while wider industry reports have found that AI-enabled adversary operations have risen considerably year on year. It shows how attackers are moving faster just as organizations are giving AI systems greater access and autonomy.
Fastly’s research shows why governance matters, with half of AI-first organizations lacking clear accountability structures, compared with 23% of non-AI-first organizations. Meanwhile, 44% say AI was directly exploited in their most recent incident, against 6% of non-AI-first peers. AI-first organizations also take 6.8 months to recover, compared with 3.9 months, and face 135% higher incident costs.
The Hugging Face incident is a reminder that these questions need to be settled before agents are given permission to act across sensitive systems. Organizations need clear ownership, limits on what agents can change and human approval for high-impact actions. Employees also need clear guidance on which AI tools are approved, what data can be shared with them and when human review is required.
Set Accountability Before the Next Incident
CISOs must be accountable for the risks within their remit, but responsibility must also sit with the teams that control the systems, budgets and decisions behind those risks. Boards need to define that ownership before an incident happens, not after. With 62% saying shared responsibility improves incident response times, there is a practical case for doing so as well as a governance one.
Accountability must be built into how organizations operate, with responsibility and authority aligned across the teams involved in managing risk. Boards must look at whether the people managing that risk had the authority and resources to do so in the first place, rather than questioning who is accountable when something goes wrong.