“An agent that’s confidently wrong is worse than one that’s slow,” said one respondent.
Adoption rates among cybersecurity teams
Published on October 13, The Enterprise Threat Intelligence Report: 2026 from Flashpoint finds that investment in agentic AI for cyber defense has risen, but actual deployment remains limited. Only a quarter (25%) of cybersecurity teams reported they are actively using agentic AI tools as part of their active cyber defense strategy. Three quarters of teams have yet to fully investigate deploying these tools: 20% said they are running a closed pilot scheme, 35% said they are examining potential adoption within the next 12 to 24 months, and the remaining 20% said they have no plans to test agentic AI for cyber defense at present.
What agentic AI is already being tasked to do
Among organizations that are using or piloting agentic AI, the report documents a range of duties assigned to those agents. Common tasks include information retrieval and alert prioritization. More active interventions reported by respondents include blocking IP addresses, updating blacklists and monitoring suspicious traffic. The report also notes that some organizations have used AI to take automated action that directly affects security operations — a step that many respondents nonetheless view as significant and, in many cases, premature.
The cyber insurance questionnaire just landed. Now what?
SOC 2, HIPAA, insurance renewals - someone has to own security strategy. Nubivance provides fractional CISO leadership without the full-time salary.
Get a security leadBarriers: trust, data leakage, auditability, regulation, cost and control
Reluctance to expand agentic AI is concrete and multi‑faceted. Respondents cited concerns about agents making wrong decisions — summed up in the warning that a “confidently wrong” agent can be worse than a slow one — and wider issues including the risk of data leakage, difficulties with auditability, regulatory requirements, cost, and questions about how much control agents should have over security workflows. Flashpoint’s founder and CEO, Josh Lefkowitz, encapsulated this caution: “The opportunity for agentic AI on the defensive side is significant, and our research shows security leaders are ready to invest. But they aren't swayed by the hype.”
Boards, budgets and the appetite to fund investigation
The report suggests a split between operational caution and executive appetite. While many security teams are hesitant to go “all in,” more senior decision‑makers appear open to funding exploration. Over half of boards and executive funding providers surveyed expect a significant budget increase to aid investigation of agentic AI for cyber defense. Specifically, 35% said they expect their cybersecurity budget to rise by over 10% in the next 12 months, and a further 15% expect an increase of up to 10%. Notably, zero respondents said they believed their cybersecurity budget was set to decrease.
How security teams, boards, and adversaries are likely to respond
- Security teams and technologists — Many will continue running closed pilots and expanding controlled use cases (information retrieval, alert triage) while resisting broad autonomous decision‑making until trust, auditability and leakage concerns are resolved.
- Boards and procurement leaders — With over half expecting significant budget increases and 50%+ signaling executive support, these groups will likely fund more pilots and vendor evaluations to determine where agentic AI demonstrably improves outcomes.
- Adversaries and threat actors — The report frames the background threat as growing AI‑enabled cyber threats to enterprise networks; defenders’ cautious stance on agentic AI means the window for operational advantage remains contested as organizations weigh capability gains against control and compliance risks.
Flashpoint’s findings leave a clear tension at the center of current planning: executives are prepared to spend, and technologists see potential, but operational trust and governance have not been settled. Security leaders are “holding agentic AI to a high bar,” in Lefkowitz’s words, and demanding evidence of where agents improve work, whether outputs can be trusted, and where human judgment must remain part of the process.
The next phase is practical: closed pilots and budgeted experiments will either demonstrate sufficient reliability, audit trails and safeguards to expand use, or they will confirm defenders’ reluctance to cede operational control. Either way, the report makes plain that investment intent exists even if widespread autonomous deployment does not — and that the questions of trust, control and regulatory compliance will determine whether agentic AI moves from pilot to frontline defender.
