This week
- A busy week following BSides Bournemouth, RISK Europe, and HSBC’s Cyber Innovation Day! Plus, I’m sure a lot of you had a hectic end to the week too (thanks, Microsoft).
Interesting stats
$740,000 in fraudulent AI usage charges racked up by attackers during Railway’s October security incident, so I’m keeping it short.
1,371 cyber incidents were reported to the UK Information Commissioner in the three months from July to September this year, beating the previous high set in 2023.
~Five~ One thing
- Microsoft has apologised to customers following an outage that caused over 24 hours of disruption to its cloud services. The incident primarily affected Redmond’s Outlook email and Teams chat services, in over 70 countries, and was not the result of a cyberattack. Microsoft says it determined that this incident was “caused by a failure in the systems responsible for establishing connections to the Exchange Online service” following standard network maintenance and upgrades. It sounds like an underlying issue caused servers handling connections to/from users’ devices to stop accepting new connections, shifting more load onto an ever-smaller pool of services, which then became overloaded in a cascade failure. A preliminary Post-Incident Report is promised within two business days, and the final version within five business days.
- There are a few things that I’d encourage all organisations to consider in the wake of the incident and realisation of how this concentration risk can manifest: Review your business continuity plan, and make sure that you have offline or redundant copies outside of (e.g.) SharePoint alone; establish a fallback channel for emergency, out-of-band communications (useful during an outage like this, or security incident where you cannot trust the confidentiality or integrity of your primary systems); make sure that you have resilient copies of key contact information (you may not have access to the Global Address List!). The fallback doesn’t need to be a complete like-for-like; it just needs to be enough to communicate key information and establish alternative methods. The same systems may be useful for communicating with managers or employees during extreme weather events, office floods, and so on.
- Also this week, Microsoft advised customers of its self-hosted Exchange Server 2016 and 2019 to consider them “permanently vulnerable”. Extended security updates have been provided since both went end of life in October 2025. Customers will need to upgrade to Exchange Server Subscription Edition (SE) to continue to receive patches.
- The US Federal Aviation Administration’s third-order national ground stop was due to a telecommunications outage, after contractor L3Harris replaced copper wire with fibre optic cables.
In brief
-
France says that Iran was behind October cyberattacks against supermarket fridges and promised European sanctions against those responsible. Iranian group Shahid Kaveh, a threat group linked to the Islamic Revolutionary Guards Corps (IRGC), has been fingered as those responsible “thanks to the remarkable efforts of government agencies,” said foreign affairs minister Jean-Noël Barrot.
-
ConnectWise is investigating reports of unauthorised access at almost 50 organisations using its ScreenConnect remote access tool. A third-party extension, RMM+, used by some managed service providers to streamline sessions, is thought to be the common factor.
-
Hoax bomb threats sent to thousands of European schools this week originated from a compromised Taiwanese university email account. The hoaxes led to evacuations at schools across Italy, and similar messages were sent to schools in Scotland, Northern Ireland, and the Republic of Ireland. Ultimately, the messages were deemed “not credible”. They follow similar hoaxes against airlines in recent weeks, similarly aimed at disruption and sowing fear.
-
Shipping giant Maersk has confirmed a ransomware attack “isolated to a local system”, as the WhiteArmor group claimed responsibility and threatened to release 100GB of information it says it stole from the company’s systems.
-
1% of Notion customers may have had “Insights” chat transcripts leaked after a supply-chain compromise at Fireworks AI. The collaborative document editor says that “any input to or output from the model that was written to the chat,” may have been exposed, but that it doesn’t believe much of the data would be attributable to individual customers. Uber was also affected by the Fireworks AI breach, and is writing to thousands of its drivers whose license information was stolen.
-
The US arm of accountancy and advisory firm Baker Tilly has suffered a data breach in September, exposing Social Security numbers, full names, addresses and other personal information stored in a third-party’s platform, and is now writing to inform affected individuals.
-
Wyndham’s Android mobile app contained a hardcoded database password that gave a security researcher access to the hotel giant’s customer data, including contact, loyalty, and authentication information.
-
SAP has fixed over 30 issues in its November security update, including four 10/10 critical vulnerabilities covering memory corruption and missing authentication in part of its core platform, and broken access control and memory corruption in the Gateway and Web Dispatcher products respectively (CVE-2026-76964, -120017, -89103, -89126; all 10/10; advisory).
-
The Apache Software Foundation has fixed a critical unauthenticated remote code execution vulnerability in its eponymous HTTP Server (CVE-2026-119618; 8.1/10; advisory).
-
Check that
.claude/settings.local.jsonis in your .gitignore file to avoid secrets and other Claude Code settings being leaked to Github. Also, if you’re vibe coding apps using SpaceXSI’s Grok Build, then the ‘random’ URL generated for your app is easy to find, and therefore your app or content accessible to anyone, because the handful of words used are part of a grok subdomain, and become available in DNS. -
German lawmakers passed legislation 308 to 190 (with 71 abstentions) allowing Germany’s intelligence agencies to break in to foreign systems, sabotage adversaries’ supply chains, and run takedowns against hostile systems.
-
Donald Trump has nominated former Palantir CISO Michael LoSapio for CISA director this week. (CISA has lacked a permanent director throughout Trump’s second term).
-
IBM and Cisco are rolling Anthropic’s Mythos model into their own security tools. Not to be left out, OpenAI has released two security products of its own, Red Team, an offensive pen testing tool, and Daybreak Defense Factory that automates “finding, validating, and fixing vulnerabilities at scale.”
And finally
- Google DeepMind has used AI to mathematically prove Signal’s cryptographic guarantees. This type of ‘formal verification’ has long been a costly and time-consuming process, typically reserved for the highest-assurance government systems, but it allows software to be proven to be free of critical vulnerabilities.