
This was highlighted very starkly during Black October, when PHP bugs and hot supermarket fridges signalled a new era of industrialised hacking. But if this weren’t enough, enterprises are also having to consider that even their own use of AI isn’t entirely safe.
“Businesses must contend with both the risks created by their own adoption of AI and the possibility that attackers will use the same advances to target them more effectively,” warns Chris Anley, chief scientist at NCC Group. Anley is referring to the fact that, aside from being targeted by AI-powered hacks, many organisations are using AI without clear oversight and governance.
A fairly substantial body of data supports this view. Most recently, London-based AI Score published a survey which found that 31% of UK organisations have experienced some kind of AI incident in the past two years. Of these organisations, 50% report the use of unapproved ‘shadow’ AI tools, while 44% reveal that confidential or sensitive data has been wrongly entered into AI applications.
Throw in regulatory breaches (21%), biased or discriminatory output (21%), and unauthorised actions by agents (13%), and the use of AI can be as fraught as the question of whether it’s really adding to the bottom line. AI Score’s research chimes heavily with earlier surveys on the use of shadow AI, including a September Deloitte study which found that UK workers are spending £958m out of their own pockets on LLM-based tools, and that only 51% of all AI-adopting employees are using tools provided by their employer.
While AI does present a variety of risks, experts affirm there are numerous policies enterprises can adopt to limit exposure. From securing greater visibility over their own tools to strengthening threat detection, such measures should reduce the downsides of AI while preserving the upsides.
The internal AI threat
What’s interesting is that, according to AI Score’s survey, 59% of organisations claim they’re doing a good job of governing AI. Yet other responses suggest otherwise, with only 24% of businesses having a complete inventory of the AI and agentic tools they use, and only 35% having full visibility of what such tools can access.
This kind of negligence may be unfortunate, since 56% of respondents indicate that requests from external parties to provide evidence of AI governance are becoming more frequent. In other words, potential clients and partners increasingly want proof that enterprises are using AI responsibly, yet it seems that many cannot provide such evidence. In response, 78% report increases in investment in AI governance, while 84% agree that robust AI governance is now a competitive advantage.
According to Alex Harland, the CEO at AI Score, enterprises investing in governance should focus on visibility. “Businesses need to understand which and where AI systems and services are being used; what data is being accessed or processed; who is doing what and, increasingly, what actions are AI agents taking,” he says. “That gives the evidence to decide where safeguards are needed, where action is required, or where progress is being made.”
Harland adds that such visibility should be continuous, especially given that tools, permissions and use cases all evolve. He also recommends that enterprises seriously consider hiring a dedicated ‘AI manager’, which isn’t necessarily an executive role, but rather a position that comes with board-level backing and assumes clear ownership over how an organisation harnesses AI.
“Every organisation using AI should have someone clearly accountable for its governance, with the authority and resources to make that accountability effective,” he adds.
Other experts have arrived at a similar view. For Martin Riley, the CTO at UK cybersecurity firm Bridewell, many businesses have deployed AI “faster than they have understood their dependency on it,” creating unanticipated risks in the process.
“Assistants are being connected to email, documents and operational processes because that is the quickest route to value,” he says. “Security teams are then left working out what those tools can reach and what they can do.”
The external AI threat
As serious as unconsidered AI adoption can be, enterprises are also having to contend with the fact that AI appears to be powering a higher frequency of cyberattacks. “The result is a more complex and fast-moving threat landscape in which traditional security approaches must evolve alongside the technology,” says Chris Anley, the chief scientist at global cybersecurity company NCC Group.
Data reinforces the suspicion that cyberincidents and attacks are increasing. The NCSC, for instance, received 1,831 reports of attacks on UK organisations between September 2025 and August 2026, up from 1,727 a year previously. Black October, which was not included in its review, suggests that 2027 might produce an even higher number.
As with Riley, Anley argues that too many businesses focus “on the productivity gains AI can deliver” as opposed to whether the underlying infrastructure, data governance and security controls are comprehensive enough to protect against internal and external threats. He also warns that AI represents a “significant” shift in cybersecurity, largely because it’s expanding the attack surface at both ends.
“The concern is that the gap between bad actors discovering a vulnerability and exploiting it could narrow, putting existing patching and incident-response processes under greater pressure,” he says. “AI does also give defenders more effective tools for detecting and responding to threats quickly, but this advantage is not automatic.”
Anley suggests that cybersecurity is becoming more of a board-level priority as a result of this shift. However, he advises that investment should keep pace with the changing threat landscape, and that organisations must remain aware that their use of models, data pipelines, AI agents and automated workflows is introducing novel vulnerabilities.
“Poorly secured deployments can expose sensitive data, create weaknesses in access controls and allow AI systems to be manipulated or misused,” he says. “At the same time, malicious actors can use increasingly capable frontier models to analyse codebases, identify software vulnerabilities and potentially accelerate their exploitation at scale.”
Riley provides a good example of a novel attack vector, explaining that Bridewell has detected exploits aimed specifically at AI assistants, with 1% of phishing attempts now containing hidden prompt injection.
“These were instructions written for the AI that summarises or acts on the email, not for the person reading it,” he says — 10% of these attempts were aimed at critical national infrastructure clients and the remainder at government bodies, while Microsoft’s prompt-injection protection blocked all of them. “The volume is low,” Riley says, “but the signal matters.”
What enterprises can do to secure themselves
As with AI Score, both NCC Group and Bridewell advocate for strong visibility of internal AI assets. “This starts with understanding where AI is being used across the organisation, what data and systems it can access, and whether appropriate controls are in place to prevent data leakage, unauthorised access and manipulation,” says Anley.
Similarly, Riley urges organisations to build inventories of their AI tools, covering embedded features, employee tools and autonomous agents. “Understand who owns each one, what data it uses and which processes depend on it,” he says, before adding that agents should be individuated and treated as identities. This means giving them defined permissions for specific tasks, as a company would do with its employees.
As for external cyber threats, Riley suggests that enterprises should be frank with themselves about the current state of their defences. “Are you releasing public-facing apps that need more DevSecOps attention?” he asks. “Are you spread across cloud environments with uneven skills and varying maturity?”
Anley also advises self-appraisal, suggesting that organisations regularly review their vulnerability management processes, while prioritising critical exposures and ensuring that patching and incident response can keep pace with attackers.
And while AI has widened attack surfaces, Anley says that these enterprises shouldn’t assume that AI tools alone will make them more secure. “AI can strengthen threat detection, accelerate investigations and help security teams prioritise vulnerabilities, but experienced professionals remain essential for validating findings, interpreting risk and making informed decisions,” he concludes.
This was highlighted very starkly during Black October, when PHP bugs and hot supermarket fridges signalled a new era of industrialised hacking. But if this weren’t enough, enterprises are also having to consider that even their own use of AI isn’t entirely safe.
“Businesses must contend with both the risks created by their own adoption of AI and the possibility that attackers will use the same advances to target them more effectively,” warns Chris Anley, chief scientist at NCC Group. Anley is referring to the fact that, aside from being targeted by AI-powered hacks, many organisations are using AI without clear oversight and governance.
A fairly substantial body of data supports this view. Most recently, London-based AI Score published a survey which found that 31% of UK organisations have experienced some kind of AI incident in the past two years. Of these organisations, 50% report the use of unapproved ‘shadow’ AI tools, while 44% reveal that confidential or sensitive data has been wrongly entered into AI applications.