At a glance
- The risk: AI is making familiar attack methods faster, cheaper, more persuasive, and available to a wider range of threat actors.
- The governance implication: Controls designed for human-paced attacks may fail when vulnerability research, impersonation, targeting, and exploitation accelerate.
- The leadership takeaway: Strengthen the decision points attackers exploit—identity, payments, releases, patching, and incident escalation—and give someone authority to stop the process.
Artificial intelligence is changing cyber risk, but not in the way the most dramatic headlines suggest. The immediate danger is not an autonomous system inventing an entirely new form of attack. It is existing attack methods becoming faster, cheaper, more convincing, and easier to repeat.
That changes a consequential leadership decision: which controls must operate at machine-assisted speed, and who can interrupt a transaction, release, or business process when the evidence no longer looks trustworthy?
The answer cannot simply be “buy more AI security.” Leaders need to identify where compressed decision time creates exposure—and where accountability still depends on a person noticing, challenging, and acting quickly enough.
The real change is attack economics
Observed threat activity shows a more grounded pattern than the idea of an all-capable AI attacker. In January 2025, Google Threat Intelligence Group reported that government-backed actors using Gemini were gaining productivity in research, troubleshooting, reconnaissance, scripting, vulnerability research, and content creation. In that dataset, Google did not observe novel breakthrough capabilities.
OpenAI reported a similar finding in February 2024 after disrupting five state-affiliated actors. Those actors used its services for open-source research, translation, basic coding, debugging, scripting, and content that could support phishing. OpenAI assessed the capability gain as limited and incremental relative to existing public tools.
Incremental does not mean insignificant. A modest productivity gain, applied repeatedly across target discovery, message creation, code troubleshooting, and campaign localization, changes the cost and volume of an operation. The strategic exposure is scale: more credible attempts can reach more people and systems before defenders adjust.
AI compresses the time available to decide
The UK National Cyber Security Centre’s assessment of AI-driven cyber threat through 2027 concludes that AI will almost certainly make parts of cyber intrusion more effective and efficient. It expects most near-term impact to come from enhancing existing tactics—reconnaissance, vulnerability research, exploit development, social engineering, basic malware generation, and analysis of stolen data—rather than creating entirely new threat vectors.
The governance consequence is decision latency. If AI helps an attacker find vulnerable systems and adapt existing techniques more quickly, a weekly remediation meeting or a multi-stage exception process may become part of the exposure. The control can exist on paper and still arrive too late.
This is particularly important for known vulnerabilities. The NCSC assesses that AI-enabled tools will increase attackers’ ability to exploit known weaknesses and further reduce the time between disclosure and exploitation. Leaders therefore need clear thresholds for emergency remediation, containment, and service restriction before a critical weakness enters the normal backlog.
Identity and approval controls carry the immediate exposure
Generative AI also improves the quality and reach of impersonation. A December 2024 FBI advisory described criminals using generated text, images, audio, and video to support social engineering, spear phishing, identity fraud, and financial schemes. The practical problem is not whether every employee can identify a deepfake. It is whether a consequential action can be approved using evidence that is easy to synthesize.
Organizations should assume that a familiar writing style, recognizable voice, realistic face, or plausible document may no longer prove identity. Payment changes, account recovery, privileged access, sensitive data release, and emergency executive requests need independent verification through a trusted channel.
Illustrative hypothetical: an employee receives a convincing voice message from a senior executive requesting an urgent payment exception. The risk is not solved by asking the employee to become a media-forensics expert. It is reduced when the process requires a verified callback, a second approver, and an explicit owner who can stop the payment without being penalized for delay.
Defence must govern decisions, not chase every signal
The wrong response is to mirror every AI claim with another AI product. Automated detection can help, but technology cannot decide which customer transaction, production release, identity event, or vulnerability exception is consequential enough to stop.
Leaders should map the decisions most likely to be accelerated or manipulated: approving payments, resetting identities, granting privileged access, deploying software, accepting vulnerability risk, and declaring an incident. For each one, the organization needs an accountable owner, independent evidence, a defined escalation threshold, and authority to pause the process.
This also means measuring whether controls operate quickly enough. Time to revoke access, contain a compromised account, patch an exploited weakness, validate a payment change, or escalate an anomalous release can be more useful than counting how many AI-related alerts were generated. The objective is not to prove that the organization uses AI in defence. It is to preserve trustworthy decisions when the adversary’s cycle becomes faster.
Leadership questions
- Which business decisions become most dangerous when an attacker can create credible content or technical attempts at greater speed and volume?
- Where do remediation, identity verification, payment approval, or escalation processes move more slowly than the threat they are meant to control?
- Who has explicit authority to stop a transaction, release, or service when the evidence may be AI-generated or the response window is closing?
Closing thought
AI does not remove the fundamentals of cyber defence. It punishes delay, weak verification, and unclear authority more quickly. The organizations that respond well will not be those with the loudest AI-security claims, but those that can make and enforce the right risk decision before the attacker’s advantage becomes an outcome.
Sources and further reading
- UK National Cyber Security Centre: Impact of AI on cyber threat from now to 2027 — an intelligence assessment using formal probabilistic language.
- Google Threat Intelligence Group: Adversarial Misuse of Generative AI — January 29, 2025.
- FBI Internet Crime Complaint Center: Criminals Use Generative Artificial Intelligence to Facilitate Financial Fraud — December 3, 2024.
- OpenAI: Disrupting malicious uses of AI by state-affiliated threat actors — February 14, 2024.
- Related Beyond1n0 analysis: Hackers Are Using AI Too and How to Reduce Your Application Security Debt.
About Gary Cheung
Gary writes about information risk, AI governance, and executive accountability. For a discussion about an AI decision or its governance, contact Beyond1n0.
