Five things from the past working week that a UK board should know about, in the order I would raise them with a chair over coffee on Monday morning. The financial regulators put cyber expectations in writing, a flaw in a twenty-year-old library is being used against Confluence and the remote-management tool many outsourced IT providers run, Microsoft's own AI agent framework was turned into a credential stealer, Europe's largest gas supplier spent four days fighting an intruder, and a British general said out loud what the National Cyber Force does. I have left out the noise. What survives has a decision attached.

1. The FCA, PRA and Bank of England have told firms what good cyber governance looks like

On 19 November the three regulators opened a joint consultation on cyber risk and resilience guidance, GC26/9 at the FCA and CP17/26 at the PRA, built around the NCSC's Cyber Assessment Framework and covering governance, identify, protect, detect, respond and recover. It closes on 26 February 2027. The numbers behind it are the point: October brought the FCA its second-highest monthly count of reported cyber incidents in five years, 52% above the previous peak, and 88% of firms in a joint Bank and FCA survey expect frontier AI to raise their cyber risk within two years. Alongside it sits a set of effective practices on AI-related cyber risk, which are not formal expectations but argue for moving from periodic vulnerability management to continuous exposure management, on the basis that AI lets attackers exploit flaws within hours or days. "Our response shouldn't always be more rules," said Nikhil Rathi. Guidance that maps to the CAF is not a new rule. It is a yardstick, and supervisors will use it.

For boards. Ask for a self-assessment against GC26/9 before the consultation closes, not after it becomes final, and ask who will respond on the firm's behalf.

2. A flaw in an old Apache library is being used against Confluence and N-able N-central

CVE-2026-90436, an unauthenticated remote code execution bug in Apache Axis 1.x rated 9.8, reaches anything still carrying that library. Rapid7 saw it exploited against Confluence Data Center on 13 November, the same day N-able reported attacks on N-central, a day after a public proof of concept appeared. Confluence is exposed where the Remote API (XML-RPC and SOAP) is switched on; Atlassian's advisory points to 9.2.28, and Cloud is unaffected. N-able's 16 November hotfix disables the legacy API and fixes seventeen further CVEs. CISA added it to the KEV catalogue on 18 November. In the intrusion Rapid7 watched, the attacker was signing in to Microsoft 365 with legacy authentication within about twelve hours. N-central is a remote management tool, and the people running it are often your managed service provider, with standing access to every machine you own.

For boards. Ask whether you, or any IT provider acting for you, run on-premises Confluence or N-central, whether the fixes are in, and how you would know if your MSP had been compromised before they told you.

3. Microsoft's official AI agent framework shipped a credential stealer

On 17 November versions 1.26.1 and 1.26.2 of Microsoft's agent-framework package on PyPI were published with a credential-stealing payload, identified by StepSecurity, OX Security and Socket. The package sees around 192,000 downloads a month. The payload runs whenever Python starts, whether or not the framework is imported, and StepSecurity's analysis shows it harvesting cloud and CI secrets, Azure managed-identity tokens and GitHub Actions OIDC tokens. PyPI quarantined 1.26.1, and the attacker simply uploaded 1.26.2. Nobody has yet explained how the publishing rights were obtained. Version 1.26.0 is the last clean release. This follows last week's stolen-token campaign through GitHub pipelines, and it lands on exactly the tooling that teams are wiring into production to build AI agents.

For boards. Ask whether your AI pilots pull dependencies straight from public registries, whether versions are pinned, and whether the credentials those pipelines hold could reach production.

4. Equinor spent four days fighting an intruder it says was not aiming at it

Equinor, one of the largest suppliers of gas to Britain and a producer on the UK continental shelf, disclosed an ongoing attack on its IT systems on the evening of 17 November. By 19 November it said the attack was still running but did not appear to target Equinor specifically; on 20 November it declared it stopped, with production normal and no sign of data stolen or altered. The Register reports that internal email was down and that Norway's security authority had already raised the national cyber risk level to four out of five. No attacker has been named. Production carried on because the operational network held while the corporate one did not, which is the separation every critical operator says it has and few test.

For boards. If your business depends on a single energy, logistics or payments supplier, ask what you would do in the four days they spend fighting something, and whether you have ever asked them how their IT and operational networks are separated.

5. Britain has said publicly that it attacks back

This week the King opened the National Cyber Force's new headquarters at Samlesbury in Lancashire. Around the opening, General Sir Rob Magowan, who leads the military's Cyber and Specialist Operations Command, told Sky News that the NCF is conducting offensive operations against Russia, China, Iran and North Korea, as Forces News reports. That has been understood for years. Saying it on the record is a signal of posture, and states that are hit tend to look for targets in return. Nothing here changes the government's advice to companies. It does change the backdrop against which a board decides how much disruption it is prepared to absorb.

For boards. Make sure your risk register treats state-aligned disruption as a live scenario, not a footnote, and that your crisis plan assumes a period where a supplier and a regulator are both dealing with the same incident.

The thread that ties this together

Most of the damage this week travelled through someone else. A library inside Confluence, a remote-management platform run by an MSP, a package published under Microsoft's name, a gas supplier with a shared risk level. Even the Irish credit unions that learned this week that payee data had been exposed through Belfast-based Wellington IT found out weeks after the October intrusion. The regulators' new guidance asks firms to evidence how they govern that exposure. The Cyber Security and Resilience Bill, which returns to the Commons on 24 November, will extend similar expectations to managed service providers.

If your most important supplier were breached on Monday, how many days would pass before you heard about it, and who on the board would be the first to know?