Cybersecurity Awareness Month finds users approving prompts on reflex as AI agents multiply

Cybersecurity Awareness Month enters its final stretch with a figure that has barely moved in a year. Verizon's 2026 Data Breach Investigations Report found the human element was involved in 62% of breaches in 2025, up from 60% in the previous edition.

“Last Cybersecurity Awareness Month, I argued that we'd moved beyond awareness and that instead organisations needed to focus on preparedness. In the intervening year, a lot has changed,” said Edwin Weijdema, Field CTO EMEA and Strategy and Cybersecurity Lead at Veeam Software. “AI stopped being an abstract tool and became part of organisations' infrastructure, with non-human identities now outnumbering human employees 109:1.”

“When an approve button gets pressed faster than a human can read the prompt, you have trained a reflex, not a security control,” said James Maude, Field CTO at BeyondTrust. “Security is often seen as introducing friction and the more hoops you make users jump through for security, the more likely they are to start ignoring them.”

“Security fatigue becomes visible when people approve prompts automatically, disable warnings or find workarounds to avoid repeated checks,” said Santiago Pontiroli, Lead TRU Researcher at Acronis. “Attackers can exploit that behaviour by flooding someone with authentication requests until they accept one. The concern is whether the user still understands and evaluates what they are authorising.”

“A user who receives an unexpected authentication request needs to recognise it as something to reject and investigate, rather than another routine interruption,” Pontiroli added. Three weeks into this year's campaign, the security leaders who responded to The Source Code pointed responsibility at the people who design products and platforms, and increasingly at the AI agents now acting on everyone's behalf.

Friction teaches people to click yes

“Anything that interrupts the user without context” creates the most friction, according to Maude. “For example, a prompt that asks 'Allow?' with little to no explanation of what, why, or what happens if you say no. So many times we are asking for a security decision from a regular user with zero information or support. Unsurprisingly, people just pick whichever option makes it go away and lets them carry on with their task.”

“Repeated sign-ins, switching apps for verification codes and confusing account recovery create friction, especially when the reason for the interruption is unclear,” said Pontiroli. “Multifactor authentication remains valuable. The problem is an implementation that repeatedly interrupts ordinary activity or leaves users struggling when they lose a device. Poor usability can encourage workarounds that weaken security, so the design of authentication matters alongside its technical strength.”

Every prompt has to earn its place

“Ask what the step actually prevents, how often, and whether the user can meaningfully make that decision,” said Maude. “If the answer is 'rarely' and 'no', then try to automate it or remove it. The best security measures are the ones the user never sees and are driven by automated, well-informed systems that make better judgements than most users.”

“Product teams should judge each security step by the harm it prevents and how people actually respond to it,” said Pontiroli. “Changing recovery details or sending money warrants stronger verification; routine activity should not require constant interruptions. Context and anomaly detection can help identify when additional verification is needed, while keeping ordinary activity straightforward.”

“Teams also need to test how people respond. Do users understand the request and make an informed choice, or simply approve it to move on?” he added. “A security step that looks effective on paper may provide little protection if its design encourages automatic acceptance.”

“Good security feels like the product working, not the product obstructing,” said Maude. “Set boundaries up front with least privilege that defaults to the minimum access an app needs, and the user never has to be the one who says no.” Pontiroli described the same goal from the user's side: “Secure defaults, phishing-resistant authentication and accessible recovery should reduce how much depends on a person interpreting every warning correctly.”

Responsibility follows control

“The party who holds the most information and has the most control should be responsible, and that is rarely the user,” said Maude. “After decades telling consumers to be more careful, we are still in the same place. We need to stop relying on the user to individually enable security and make things secure by default. This is the only way that security scales.”

“Consumers cannot independently inspect firmware, cloud infrastructure or every process through which a company shares their information,” said Pontiroli. “A stronger password on their account cannot prevent a company from disclosing their data through a compromised internal process. Users still have a role: protecting accounts, keeping software updated and checking unexpected requests through an independent channel. Those habits should be supported by systems designed to limit the consequences of mistakes and compromised trust.”

“Cyber fraud puts that trust at risk, along with a family's lifetime savings or a small business's working capital,” said Manish Mimani, Founder and CEO of Protectt.ai. Mimani was speaking about the UAE and Saudi Arabia, where national digital ambitions depend on people trusting the services they use every day, though the exposure he describes applies in any market where banking and payments now run through a phone.

“Advanced artificial intelligence is enabling attackers to generate hyper-realistic identities, convincing messages, and sophisticated scams at unprecedented scale,” Mimani added. “No single cybersecurity enterprise, banking institution, or fintech player can solve this crisis in isolation.”

The newest users never see a prompt

“This year, I believe organisations need to shift back to awareness, so they can understand just how fundamentally AI has changed the structure of their digital estate,” said Weijdema. Verizon's report tracks that change from the employee side: regular AI use on corporate systems rose to 45% from 15% a year earlier, and 67% of users accessed AI services through non-corporate accounts on company devices.

“AI safety is not only a model problem,” said Frank Ruge, SVP EMEA at Infoblox. “Agents are systems with access to data, networks and services. The control question is simple: what can they reach, what access do they have, who owns them, and how do you stop them?” Infoblox positions DNS as an early point of visibility and enforcement, while acknowledging that containing AI also requires identity controls, segmentation, egress restrictions, telemetry and tested response workflows.

“Every autonomous agent needs clear boundaries around the services it can discover and contact, with policy and visibility at the network layer, not assumptions that the application boundary alone will hold,” Ruge added. He pointed to the UAE, which is moving quickly to put AI into high-value public and private-sector services, as one market where resilient and accountable infrastructure has become essential.

“Cybersecurity leaders want autonomous threat hunting that keeps working while their teams sleep, and policy-driven decisions that run in real time across their systems,” said Alexandre Depret-Bixio, SVP International at Anomali. “Most organisations struggle with the fidelity of their data, and if you can't trust your data, you can't trust the decisions AI agents make on your behalf. You end up with confident but bad decisions, and in cybersecurity, that risk is untenable.”

Compliance snapshots expire on contact

“Too many still rely on outdated snapshot metrics, like compliance status, to reassure themselves of their resilience,” said Weijdema. “In an environment where overprivileged AI agents are constantly accessing data and acting at machine speed, these metrics are outdated from the very second they are measured.”

“Cybersecurity Awareness Month should serve as a catalyst for turning security fundamentals into operational discipline,” said Ray Heffer, Field CISO at Veeam Software. “Organisations need to enforce the basics: strong authentication, phishing awareness, timely patching, least-privilege access, and tested recovery. With the use of AI, that discipline must extend to clear visibility into where data lives, how it flows, who or what can access it, including AI agents, and whether clean, trusted data can be recovered when something goes wrong.”

“Autonomy earns trust one governed decision at a time, with human approval and audit where it matters,” said Depret-Bixio. The human approval he describes is the same checkpoint Maude and Pontiroli warn has decayed into a reflex for ordinary users, which means the prompt will only protect anyone if it carries the context, the stakes and a clear way to say no. With the month's final days ahead, Depret-Bixio offered its closing test: “This Cybersecurity Awareness Month, the lesson is clear: the near-term future of AI will be judged by whether we can stand behind every decision it makes.”