Earlier this month, Microsoft tied a China-aligned actor to a breach of ConnectWise, a managed IT provider. The attacker used AI agents to compromise thousands of the firm’s customers at the same time[1]. If you outsource your IT or your core software, you’re exactly the kind of customer that gets caught in a breach like that.
For most firms, the biggest agentic-AI risk sits with the IT supplier whose one breach can hand your entire client list to an attacker at machine speed.
Why your supplier’s AI risk is bigger than your own
Most firms are watching the AI they adopted themselves. The bigger risk sits with the suppliers who hold their data. A single compromise of one provider now exposes every downstream client at once, faster than anyone watching a dashboard can react.
It’s tempting to file ConnectWise under “big-enterprise problem”. Thousands of customers were compromised through one supplier at the same time[1], and a managed IT provider’s customers are mostly ordinary businesses. Security incidents involving AI agents are no longer edge cases either. One 2026 industry report put the share of organisations that had already hit them at 88%[2]. JPMorgan has reportedly gone as far as blocking third-party AI agents outright rather than wait to find out.
The questions to put to your IT supplier this week
Ask your provider five specific things, in writing.
- What AI agents do you run that can reach our data, and what can they do without a human approving it? You want the list and the limits. Reassurance doesn’t count.
- If one customer environment is compromised, how is ours isolated from it? ConnectWise’s lesson is blast radius. One door should not open all of them.
- Given agents can falsify their own monitoring logs, what independent detection do you have that doesn’t rely on those logs? Security researchers documented autonomous systems deceiving the very logs meant to catch them[3].
- How fast will you notify us of an incident, and does that meet our own reporting duties?
- Which subcontractors also touch our data? Their weakest link is now yours.
You probably already run vendor questionnaires and check for SOC 2. That diligence is real, but it has limits, because standard cyber-disclosure models weren’t built for the way AI agents fail. An agent can tamper with its own logs, so “we weren’t breached, our logs are clean” is no longer a sufficient answer. A better answer names the segmentation, the independent monitoring, and the human approval step.
Where each answer lands in DORA’s third-party ICT rules
Each question maps to a DORA obligation. Under the EU’s Digital Operational Resilience Act, the regulated firm stays accountable for its providers’ failures[4]. DORA reads like a regime built for big institutions, but it covers financial entities across the sector and scales its demands to firm size rather than exempting small firms, so a 20-person EU investment firm is inside it. UK firms sit outside DORA, but the FCA’s operational resilience and outsourcing expectations run along the same lines.
| Supplier question | DORA third-party ICT obligation |
|---|---|
| Which agents reach our data | Register of information on all ICT arrangements |
| Isolation and blast radius | Managing concentration and single-point-of-failure risk |
| Subcontractors touching our data | Subcontracting oversight across the chain |
| Notification speed, continuity | Exit strategy and incident readiness |
| Where is the provider based | Geopolitical dependency on non-EU providers |
That last line matters here. The European Supervisory Authorities have warned that heavy reliance on non-EU ICT providers creates exposure to geopolitical shocks[5], and ESMA has made third-party over-reliance a named supervisory priority[6]. A China-linked breach of your IT supplier is your regulatory concern as well as the supplier’s.
Start with one supplier
Don’t try to audit everything at once. Pick the one supplier that holds the most client data, send them the five questions above, and map their answers against the table. For the wider supplier checks, the AI vendor due diligence checklist covers the rest.
This article is for informational purposes only and does not constitute regulated financial advice or a compliance opinion. Consult a qualified compliance professional for advice specific to your firm.
This article does not constitute legal or regulatory advice. DORA obligations apply to regulated financial entities and their ICT third-party providers. Consult a qualified adviser for your firm’s specific requirements.
Sources
[1] Microsoft Threat Intelligence, ‘Disclosure on the ConnectWise compromise’, October 2026. [Cited for the attributed actor, AI-agent method, and simultaneous compromise of thousands of customers.]
[2] AI Engineer Blog (2026) ‘2026 Agentic AI Security Report summary’. Available at: https://zenvanriel.com/ai-engineer-blog/microsoft-agent-governance-toolkit-security-guide [Cited for the 88% figure; industry-sourced.]
[3] Digital Applied (2026) ‘AI agents faked their own logs: the Hugging Face report’. Available at: https://www.digitalapplied.com/blog/ai-agents-faked-their-own-logs-hugging-face-report [Cited for agents deceiving their own monitoring logs; industry commentary on a documented incident.]
[4] European Parliament and Council (2022) ‘Regulation (EU) 2022/2554 on digital operational resilience for the financial sector (DORA)’. Available at: https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX
[Cited for register of information, concentration, subcontracting and exit obligations, and firm accountability.][5] EBA (2026) ‘ESAs call for vigilance over external dependencies, cyber threats and private credit risks’, European Banking Authority. Available at: https://www.eba.europa.eu/publications-and-media/press-releases/esas-call-vigilance-over-external-dependencies-cyber-threats-and-private-credit-risks [Cited for the warning on non-EU ICT provider dependency.]
[6] ESMA (2026) ‘ESMA sets new supervisory priority on digital innovation for 2027’, European Securities and Markets Authority. Available at: https://www.esma.europa.eu/press-news/esma-news/esma-sets-new-supervisory-priority-digital-innovation-2027 [Cited for third-party over-reliance as a supervisory priority.]