AI Security Has an Accountability Problem: What Our Latest Data Says About Who Owns the Call
When an AI-assisted tool gets a security finding wrong, who takes the blame? In HackerOne's Security Research Report 2026, Exposure Debt: The Next Challenge in Cybersecurity, 38.7% of security leaders named the CISO as the role most likely to be held responsible. Only 9.9% said responsibility would be shared through a formal accountability framework.
The survey of 111 security leaders at organizations with more than $100 million in annual revenue also shows what worries them most. Asked about delegating security decisions to AI, 47% named accuracy as their top concern, specifically false negatives: real vulnerabilities an AI tool rejects or misjudges, which then go unaddressed. Without a process to catch them, those misses can go unnoticed.
Taken together, the numbers point to a gap. Organizations are handing more security decisions to AI, and most have yet to decide who answers for the outcome, even as AI opens new attack surfaces across the business.
Key AI Findings from the 2026 Data
- 39% of security leaders say the CISO is most likely to be blamed when an AI-assisted tool gets a finding wrong.
- 10% say responsibility would be distributed through a formal accountability framework.
- 47% cite accuracy, specifically the risk of false negatives, as their top concern about delegating security decisions to AI.
- More than a third of security leaders report that executive leadership only partially understands the difference between discovery, validation, and remediation.
- Valid AI vulnerability reports grew 222% in two years, and Excessive Agency reports tripled.
Who Is Accountable When AI Gets a Security Decision Wrong?
After the CISO, security leaders most often named security operations leaders (20.7%) and AI tool vendors (15.3%). Just 1.8% said no one owns the call today.
Default accountability carries real risk. Without a framework, there is no shared standard for when AI can act on its own, when a person must review its output, or how a missed finding gets traced back to its source. The CISO absorbs the consequences of decisions made by systems they may never have selected, configured, or approved, while the teams that deploy those tools have no defined role in the outcome.
The problem compounds at the top. In more than a third of organizations, executive leadership only partially understands the difference between discovery, validation, and remediation. When executives can't see where in that chain an AI tool made a decision, assigning ownership for a bad one becomes harder still.
For some security leaders, the fix is to make that ownership explicit and complete. As Kyle Barry, CISO, CLEAR explains:
“Accountability for any security program or tool should squarely fall on the security team… They need to own the tools, the process, and the experience end to end.”
Why Are Security Leaders Cautious About Delegating to AI?
Accuracy tops the list. Beyond the 47% who worry about false negatives, security leaders point to regulatory exposure (41%), loss of human oversight (41%), and adversarial manipulation of AI agents (37%).
The focus on false negatives is well placed. A false positive costs an analyst time to review and dismiss. A false negative costs visibility, because a real vulnerability never reaches the people who could fix it.
The pressure to delegate is still real. Security leaders report an average of 30 hours to validate a critical finding, and nearly 70% say validated findings are entering their backlog faster than their teams can remediate them, a core driver of exposure debt. AI promises to compress that work. Without a defined owner for AI judgments, though, speed can introduce a new kind of exposure.
How Is AI Expanding the Attack Surface?
The accountability question extends beyond the AI used inside security programs. Valid vulnerability reports involving AI systems grew 222% over two years, and one category speaks directly to the accountability gap.
Excessive Agency vulnerabilities occur when AI systems can act beyond their intended limits. Reports of them on the H1 Platform tripled, from a small base, from 2025 to 2026. As models connect to more tools, data sources, and business workflows, the question of who approves an AI system's actions applies to production agents as much as to security tooling. Security leaders see it coming: 60% name AI-accelerated attacker capabilities as a top emerging challenge, and 42% name agentic AI and MCP security.
Want to learn more about AI and accountability in the 2026 data?
Watch our on-demand webinar, The Vulnerability Is Validated. The Clock Is Running. Now What? to explore how AI is reshaping vulnerability discovery and accountability, and what security leaders can do to close the gap between finding risk and fixing it.
How Can Organizations Keep the Blame Question from Going Unanswered?
As AI tools take on more security decisions, the organizations using them need frameworks and processes that settle accountability before something goes wrong. At a minimum, that means a clear way to challenge an automated decision and a person responsible for reviewing it.
On the H1 Platform, that principle takes the form of a defined escalation path. Automated systems help manage report volume, while customer program teams set severity and bounty amounts, with AI agent input treated as advisory. When a researcher disputes a closure or an amount, Mediation brings in a person to review it, and where a report warrants another look, HackerOne works with the program to reassess it. How those systems work is documented in Responsible AI at HackerOne.
What Should Security Leaders Do About AI Accountability?
The report's guidance is direct: define AI accountability before an incident forces the issue. That means formally setting the limits of automated tool decisions, where human review is mandatory, and who owns escalation. The NIST AI Risk Management Framework offers a starting structure through its four functions: Govern, Map, Measure, and Manage.
In practice, the work starts with four questions every executive team can answer together:
- Which security decisions can AI tools make on their own, and where is human review mandatory?
- Who owns escalation when an engineer or researcher challenges an AI-supported decision?
- How are false negatives detected, measured, and reported to executive leadership?
- Which AI systems in production have been tested the way an attacker would test them?
Written answers turn default blame into defined ownership. They also give the CISO something fewer than one in ten security leaders report having today: an agreement, approved above the security team, on how AI risk is governed.
When an AI-assisted tool gets a finding wrong, someone will be asked to answer for it. Organizations that decide who in advance get to answer on their own terms.
Next Steps: From AI Adoption to AI Accountability
The 2026 Security Research Report shows how AI is reshaping vulnerability discovery, the attack surface, and the accountability questions that come with both. Explore the findings and what they mean for your security strategy.
Download the 2026 Security Research Report Get the full data on AI accountability, AI vulnerability trends, and exposure debt.
Watch Our Webinar Hear security experts unpack the findings and discuss how organizations can govern AI-supported security decisions.
