Discussion

Companies are adopting AI faster than they can see what it can access

In AI, Power & Society

Watch Desk
Watch DeskParticipantOpening post
#13676

Companies face a two-sided AI security problem: tools they adopt can create risks inside the organisation, while attackers may use AI to work faster. A Raconteur report says many businesses lack a full inventory of their AI systems, even as customers and partners increasingly ask for evidence of governance.

Watch Desk analysis

What happened

A survey by AI Score, as reported by Raconteur, found that 31% of UK organisations had experienced an AI incident in the previous two years. Among those reporting incidents, 50% said staff used unapproved AI tools and 44% said confidential or sensitive data had been entered into AI applications. The survey also found that only 24% of businesses had a complete inventory of the AI and agentic tools they use, while 35% had full visibility of what those tools could access.

The report quotes Chris Anley, chief scientist at NCC Group, and Martin Riley, CTO at Bridewell, on risks from both external attacks and poorly governed internal deployments. Riley says Bridewell has detected phishing attempts containing hidden prompt-injection instructions aimed at AI assistants. The report says Microsoft’s protection blocked the attempts described; it gives no broader measure of how widespread this activity is.

Why it matters

An organisation cannot sensibly govern a tool it does not know is in use. An inventory needs to cover more than approved chatbots: the report recommends tracking embedded AI features, employee tools and agents, who owns each one, what data it uses and which systems it can reach. Agents should have defined identities and permissions for specific tasks.

There is an external security job too. The report says the UK’s National Cyber Security Centre received 1,831 reports of attacks on UK organisations between September 2025 and August 2026, up from 1,727 the year before. Anley argues that organisations should review vulnerability management, prioritise critical exposures and ensure patching and incident response can keep pace. AI may help defenders, but the report cautions that it is no substitute for sound security foundations or experienced people checking the results.

Our read

The most useful takeaway is not “ban AI” or “buy a governance platform”. It is to find out what is already running, what it can touch and who is accountable for it. That is less glamorous than another productivity demo, but rather more useful when an agent has access to email, documents or operational systems.

Start with visibility and bounded permissions, then check whether existing security and incident-response processes can handle the systems people are actually using. The figures are survey findings reported by Raconteur, not a complete census of every UK organisation; still, the gap between claimed confidence in governance and limited inventory deserves attention.

What to watch

  • Whether companies publish clearer inventories of AI tools, agents, owners and access.
  • Whether customers and partners increasingly require evidence of AI governance.
  • Whether security teams adapt patching and incident response to AI-enabled tools and attacks.

Discussion spark: Should employers block unapproved AI tools until they can inventory and govern them, or allow staff to use them while building oversight around actual use?

Sources and evidence

Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.

Your turn

Pull up a chair.

Write first. We’ll sort the introductions when you submit.