FCA authorised · FRN 724952 0117 325 0027Quote & buy →
Aipedia: IT consultants and developers
If an AI agent you deploy deletes data, sends wrong messages or breaches data protection law, the client will usually blame failures in your design, permissions or oversight. Agents act rather than advise, so their errors land directly in live systems.
Part of: AI in IT consultants and developers
In short
Agentic AI can read email, update records, run scripts and call other systems with little human input. That autonomy is the risk: a misread instruction or a manipulated input can trigger actions at scale. On 8 October 2026 the ICO opened a call for evidence on agentic AI and warned that "the fact AI agents act with autonomy is not an excuse for poor compliance". The AI Cyber Security Code expects humans to stay accountable. Technology PI usually responds to negligent design or deployment, subject to the wording, while data loss and system damage may also engage cyber cover.
Consultants now build agents that triage inboxes, raise and close tickets, reconcile records, provision user accounts or run routine maintenance. The agent is typically a large language model connected to tools through APIs and service accounts, often with the client's credentials. The consultant designs the workflow, sets permissions and decides when a human must approve an action. The client then relies on the agent to act unsupervised for much of the time.
The ICO's announcement of 8 October 2026 extended its scrutiny to AI agents, with a call for evidence closing 20 November 2026. Its guidance pipeline lists an agentic AI guidance consultation for October 2026 and a final version in Spring 2027, and a statutory code on AI and automated decision-making is forthcoming. Until then, existing UK GDPR duties apply to any personal data an agent touches.
The AI Cyber Security Code of Practice is relevant at several points: principle 4 (enable human responsibility for AI systems), principle 6 (secure your infrastructure), principle 9 (testing before release) and principle 12 (monitor system behaviour). For BCS members, clause 2.6 (avoid injuring others through negligent acts) and 3.1 (due care and diligence) apply to the way permissions and safeguards are designed.
Illustrative scenario: a consultancy deploys an agent to tidy a client's customer database by merging duplicates. The agent is given write access to the whole database rather than a staging copy. A loosely worded rule causes it to merge distinct customers with similar names, overwriting contact and order history for thousands of records before anyone notices. The client faces complaints, a possible data protection breach and a costly restoration, and alleges the consultancy was negligent in granting broad permissions without approval steps or rollback.
Other runaway patterns include agents looping on a task, emailing customers in error, or following instructions hidden in an incoming document. That last one, prompt injection, is a known risk for any agent that reads untrusted content.
Technology PI is usually the policy that responds when a client alleges your design, configuration or advice was negligent, including the client's costs of putting things right, subject to the wording. Cyber insurance typically focuses on security incidents, including your own and sometimes clients' data restoration and incident response costs. An agent failure may sit across both, especially where an external attacker exploited the agent. Contract terms matter too: an indemnity for "all losses caused by the agent" may exceed what PI will pay.
Our pages on AI agents insurance and PI versus cyber for IT consultants explain where the line usually falls. There is no compulsory PI for IT consultants; limits come from client contracts.
Apex arranges PI and cyber cover for IT consultants who build and deploy AI. A broker reviews how both policies would respond to an agent failure. Related guides: building AI chatbots, AI in managed services and the IT consultants AI hub.
Both can be. The client is usually responsible to third parties and regulators as the organisation running the agent. It may then claim against the consultant if the design, permissions or testing were negligent or did not meet the contract. Liability caps and scope of work will shape the outcome.
On 8 October 2026 the ICO opened a call for evidence on agentic AI, closing 20 November 2026, and warned that autonomy is not an excuse for poor compliance. It plans agentic AI guidance, with a final version due in Spring 2027, and a statutory AI and automated decision-making code.
If the client claims restoration costs from you as damages for negligent work, PI commonly responds, subject to the wording and limit. Your own costs of restoring systems, or losses from a cyber attack via the agent, are more often cyber insurance matters. Check both policies together.
If deploying agents with write access to client systems is a material part of your services, it is sensible to describe it accurately on the proposal form and at renewal. Failing to disclose a material change in your business could affect how a claim is handled, depending on the policy terms.
Tell us about your firm and the work you do. A broker will review how your current PI treats AI-assisted work, or arrange new cover. No fee for a quote. Or call 0117 325 0027.
Start a PI quote Request a callbackApex Insurance Brokers Limited is authorised and regulated by the Financial Conduct Authority. Registered in England and Wales, company number 07014570. This page is general information about how AI is used in the profession and how insurance may respond. It is not legal, regulatory or insurance advice on your circumstances, and it does not guarantee that cover will be available or on what terms.
