The tools were new. The failures were not.
Edition 9 of Cyber Governance in the AI Era, first published on LinkedIn on 4 November 2026.
In brief
Black October asked one question: are we affected?
It sounds simple. Regulators were not prepared to leave the answer to chance. In the week of 21 October, New York's financial regulator told every institution it supervises to determine whether it used the vulnerable software at all, and Britain's National Cyber Security Centre advised every organisation using it to assess its exposure.
The month brought critical flaws in PHP found with the help of AI, AI-assisted intrusions, security incidents at platforms developers use to build AI products, and the theft of about 6.4 billion dollars from Binance, which the blockchain analytics firm Elliptic calls "almost certainly the largest single theft of any kind in history." Some cybersecurity coverage began calling it Black October. The technologies differed. The weaknesses underneath them were familiar: exposed interfaces, unlisted components, misplaced credentials, unchecked signatures.
This edition measures the month against four clocks that any board can ask management to time.
Time to know. Are we affected?
Time to decide. Who has the authority to act?
Time to fix. How quickly can we patch or isolate?
Time to tell. When do regulators, customers and counterparties hear from us?
The proposition
My position, stated so it can be tested against the record: the incidents that cost institutions the most over the next twelve months will begin, as October's did, with something the institution could have known about. AI is shortening the time between that weakness and its exploitation. In October's evidence it was rarely the weakness itself.
Board takeaway. Before the end of the quarter, ask management for four numbers and one name. How many hours it would take to answer "are we affected?" for a critical flaw in a common component. How many hours to patch or isolate it. Who holds the authority to take a service offline, by name, and whether they have ever used it. What independent check sits between any request and an authorisation that moves money, releases data or changes control. And how long it would take to tell the people who need to know.
What October showed, and what it did not
It helps to separate three things the coverage ran together.
First, AI found the flaws, and it found them on the defenders' side. On 21 October the maintainers of PHP, the language behind 69.3 percent of websites whose server-side language is known, released fixes for two critical flaws that let an unauthenticated stranger run commands on a web server. The PHP Foundation credited Calif, OpenAI and Wiz, among others, for the research and responsible disclosure. Wiz says its researchers discovered the first flaw using its own platform and two frontier models and reported it to PHP on 24 August. That is the good news, and it is real. The bad news sits at the other end of the disclosure. Amazon's threat intelligence team saw an exploitation attempt against its honeypots on 17 October, four days before the fixes were public. CVE identifiers leaked on 20 October and the release was moved forward a day. CISA added both flaws to its catalogue of known exploited vulnerabilities on 26 October and gave federal agencies until 29 October to fix them.
Second, AI is now verifiably in attackers' hands, though not in the form the headlines suggested. Microsoft reported on 21 October that a China-aligned group it calls Storm-0940 had used ARTEX, "an open-source penetration testing framework that can use any AI model," against hundreds of organisations, and that in multiple instances the time from initial targeting to privileged access was under three hours. Microsoft was careful on both sides: "the majority of the targeting wasn't successful," yet the group "achieved privileged cloud persistence at many organizations in a brief time." On 26 October the research firm Dream described a group that had built its own PHP exploit, chained it with a Linux privilege flaw and rooted 25 servers, including firewalls at a US government-affiliated organisation. The complete chain, Dream says, runs in under fifteen minutes. On the attackers' infrastructure it found Claude Code installed and an offensive AI platform configured with Chinese models. Dream is precise about what that proves: it has "no direct evidence that the AI tools were specifically used to write the exploit," and the operation "does not appear to be a fully autonomous nation-state attack." And the chain, in Dream's words, started with "a management interface that should never have been exposed to the internet." In the same weeks, at least nine platforms that developers use to build AI products disclosed security incidents. Wiz Research found that the public reports all describe targeted exploitation hunting stored environment variables, where applications keep their secrets, "with a focus on those granting AI access."
Third, the predicted flood did not arrive on schedule. Axios, reviewing the month on 28 October, reported that millions of attempts against the PHP flaws had produced few successful attacks, partly because the flaws proved unusually hard to exploit, and partly, it judged, because defensive work at many companies appears to have helped. Its comparison was Y2K, which, it noted, "was averted through massive, coordinated effort, not luck." Britain took 32 government services offline, and its minister told the House of Lords that was "largely from precautionary measures."
Read together, the record supports a sober conclusion. AI widened the funnel at both ends. It found more flaws, and it helped some attackers move faster through the openings institutions had left. In the incidents with the clearest evidence, those openings were old ones.
The theft that needed no AI
The month's largest theft is the cleanest illustration.
On the evening of 29 October, Binance's hot wallets were drained. Binance puts the loss at about 6.3 billion dollars; Elliptic, tracing the funds across eleven blockchains, puts it at 6.43 billion. On 31 October the FBI attributed the theft to North Korea's TraderTraitor group, the unit it named for the 1.5 billion dollar Bybit theft in February 2025.
Binance's account is incomplete, and it says so. It first said the incident "may involve a compromised DevOps engineer's laptop," then that investigators had "uncovered additional points of breach" and found "no evidence that wallet private keys were leaked." Co-chief executive Richard Teng said the delegation the attacker used "did not pass through our normal approval controls."
There is no public evidence that AI played any part. The FBI's notice does not mention it. Elliptic, which looked specifically for AI agents in the laundering, found that question "cannot be determined from the on-chain data." Its lesson applies well beyond crypto: as with Bybit, "it was the infrastructure and people around the wallets that were compromised, not the cryptography."
The response is instructive too. Tether froze the attacker's main wallet 37 minutes after the stolen funds arrived, and Elliptic has verified 2.85 billion dollars frozen or seized. Binance is covering customer losses from a reserve that existed long before the night. Neither the issuer relationships nor the reserve could have been improvised.
Most institutions run systems that can be instructed to pay, transfer, release or sign. The board's question is whether anything independent of the requester stands between the request and the signature.
Four clocks
Each clock can be measured before the next incident starts. October showed what happens when they run at different speeds.
Time to know. How long does it take to answer "are we affected?" The attackers Dream tracked used an internet asset search service to fingerprint about 19,000 internet-facing firewalls before they chose their targets. They had an inventory of exposed systems. The Cayman Islands Government could say publicly on 22 October that its two main public platforms do not run PHP, and that every other system was being assessed individually. That sentence is only possible for a government that knows what it runs. New York's regulator extended the question past the institution's own walls, telling firms to check whether PHP is used "across your organization" and to coordinate with third-party service providers to assess their exposure. The third-party clause is the one to read twice.
Time to decide. Who can take a service offline, and will they? CISA told organisations that could not patch to take affected systems offline. New York's regulator said web application firewalls "should not be used as a substitute for patching or isolation." Britain took 32 services down rather than wait. The lawyers at Debevoise, writing on 31 October from inside the incident-response market, put the governance point plainly: "you need to ensure you know who has the authority to do that and make it clear to them that they need to act." Isolation is a business decision with a price. A board that has not priced it in advance will have it priced for them during the incident.
Time to fix. Microsoft measured under three hours from targeting to privileged access. Dream measured under fifteen minutes for the full chain once an exposed interface was found. At Binance, 2.53 billion dollars left in a single transaction. Against that, the US federal deadline after the flaws were confirmed as exploited was three days, and that was considered urgent. Few organisations will patch at machine speed. What they can know is which systems would be patched in hours, which in days, and which cannot be patched at all and should therefore not face the internet.
Time to tell. The best example of this clock came from a small state. On 22 October Jamaica Electronic Transfer Services Limited (JETS), which operates the country's MultiLink debit network and automated clearing house, disclosed suspicious activity in part of its internal administrative systems, and the Bank of Jamaica followed with a statement the next day. Jamaica's technology minister, Dr Andrew Wheatley, did not treat early disclosure as an embarrassment. He opened a formal inquiry, and he also said this: "Early, transparent disclosure is exactly the conduct we want to see from operators of critical systems, because it allows the whole country to respond together." Compare that with the incident at the centre of Edition 8, where a developer's agent entered an Australian government system in June and the government learned of it in September.
None of these clocks needs AI to fix. Each of them gets more expensive the longer it runs, and AI is what is making them run faster.
What two small states got right
For a small state, Black October carries a hard lesson and an encouraging one.
The hard lesson is about inventory. The software at the centre of the month was not exotic. It runs government portals, ministry websites and the management interfaces of security appliances bought from abroad. A small administration rarely builds these systems itself. It buys them, often through a contractor, and the knowledge of what is installed where can leave with the contract. The first question of the month, "are we affected?", is hardest to answer in exactly the places with the fewest people to answer it.
The encouraging lesson is that the best public responses I read this month came from the Caribbean. Jamaica's Office of the Prime Minister issued directives to every ministry, department and agency to apply the updates, report the status of their systems and report any sign of compromise to the Cyber Incident Response Team Division of the Office of the Prime Minister, which would monitor compliance. "We did not wait to find out whether these vulnerabilities would be used against us," Dr Wheatley said. Cabinet approved a National Cybersecurity Coordination and Assurance Council with a 24-month mandate, including support for a financial sector CIRT led by the Bank of Jamaica and the Financial Services Commission. Cayman, for its part, used the moment to warn residents about scams that exploit news of security problems.
Neither government needed a frontier laboratory. They needed a list, a reporting line, a named authority and the willingness to speak early. That is institutional discipline, and both governments showed it does not depend on size.
Signal of the month
Jason Clinton, deputy chief information security officer at Anthropic, wrote on 22 October: "Security leaders, never let a crisis go to waste: please go to your boards and ask for help right now. Not for ai, for the basics." He expected the pressure to last about eighteen months. From inside one of the companies building the models, that is a request for the fundamentals October showed were missing. Watch for Binance's promised post-incident report, which should explain how one laptop led to valid signatures on so many wallets in a single evening, and for the report Dr Wheatley has requested from the Bank of Jamaica and JETS.
Five questions for your next board meeting
If a critical flaw in a common component were announced tomorrow, how many hours would it take us to say, with evidence, whether we are affected, including through our vendors?
Which of our management interfaces, consoles and administrative systems can be reached from the internet, and who decided that they should be?
Who, by name, has the authority to take a service offline during an incident, what would it cost us per day, and has that decision ever been rehearsed?
What independent check stands between a request and any signature, payment or release of data that our systems can make, and how quickly would an anomaly be seen?
When something goes wrong, how long before we tell our regulator, our customers and our counterparties, and who decides?
The mandate
Know what faces the internet. Know what runs inside it, including at your suppliers. Know who can shut it down, and rehearse the decision. Put an independent check between every request and every authorisation that moves money, releases data or changes control. Decide your disclosure clock before an incident decides it for you.
AI is shortening the time institutions have to make up for weak fundamentals.
The executive question
One group of attackers in October fingerprinted about 19,000 exposed firewalls before choosing its targets. Could your institution produce its own list before someone else uses theirs?
The tools were new. The failures were not.
Sources
Wiz Research, "PHP2Shell: Critical Remote Code Execution in PHP," 21 October 2026: wiz.io/blog/php2shell-php-rce
Tenable, "PHP2Shell (CVE-2026-109618, CVE-2026-109619): Frequently Asked Questions," October 2026 (Amazon pre-disclosure observation, CVE leak, CISA KEV dates, W3Techs share): tenable.com/blog/php2shell-cve-2026-109618-cve-2026-109619-frequently-asked-questions-about-critical-php-remote
CISA, "Critical Vulnerabilities in PHP Require Immediate Patching," 21 October 2026: cisa.gov/news-events/alerts/2026/10/21/critical-vulnerabilities-php-require-immediate-patching
New York State Department of Financial Services, Industry Letter on PHP patching, 22 October 2026: dfs.ny.gov/industry_guidance/industry_letters/il20261022-cyber-php-patching
Microsoft Threat Intelligence, "Storm-0940 uses ARTEX agentic AI in ConnectWise supply chain compromise," 21 October 2026, updated 26 October: microsoft.com/en-us/security/blog/2026/10/21/storm-0940-uses-artex-agentic-ai-connectwise-supply-chain-compromise
Dream Research, "Chained PHP2Shell Exploit Roots 25 Servers, Including US Government Servers," 26 October 2026, updated 29 October: dreamgroup.com/blog/chained-php2shell-exploit-roots-25-servers
The Register, "PHP2Shell puts dozens of UK government services out of service," Carly Page, 22 October 2026
Axios, "AI's cyber scare delivers a Y2K lesson," Sam Sabin, 28 October 2026: axios.com/2026/10/28/cybersecurity-black-october-ai-hacks
CNBC, "AI's 'Black October': Critical PHP flaws set off global security scramble," Ashley Capoot, 23 October 2026
Wiz Research, "Black October for AI developer platforms," 30 October 2026: wiz.io/blog/black-october-for-ai-developer-platforms
Debevoise and Plimpton Data Blog, "Black October in Cybersecurity," 31 October 2026: debevoisedatablog.com/2026/10/31/black-october-in-cybersecurity
Binance, "Security Incident Update #2," 30 October 2026: binance.com/en/support/announcement/detail/d3306d9dca5243548c813b605e1d1751
FBI, "North Korea Responsible for $6 Billion+ Binance Attack," 31 October 2026: fbi.gov/investigate/cyber/alerts/2026/north-korea-responsible-for-6-billion-binance-attack
FBI, PSA I-022625-PSA, "North Korea Responsible for $1.5 Billion Bybit Hack," 26 February 2025
Elliptic, "The largest theft in history: North Korea likely behind the $6.4 billion Binance hack," Tom Robinson, 30 October 2026, updated 1 November: elliptic.co/insights/binance-largest-hack-in-history
Richard Teng, video statement on X, 30 October 2026
Office of the Prime Minister of Jamaica, "Minister Wheatley outlines Government response to JETS cyber incident and critical web software vulnerabilities," 23 October 2026, updated 26 October: opm.gov.jm
Bank of Jamaica, "Statement on cyber incident at JETS Limited," 23 October 2026: boj.org.jm
The Gleaner, "Wheatley launches inquiry into JETS cyber breach," 26 October 2026: jamaica-gleaner.com/article/news/20261026/wheatley-launches-inquiry-jets-cyber-breach
Cayman Islands Government, "Cyber Security Readiness and PHP Vulnerabilities," 22 October 2026, including the UK National Cyber Security Centre alert of 21 October: gov.ky/w/cyber-security-readiness-and-php-vulnerabilities
Dr. Inshan Meahjohn is Founder/CEO of DAG (Digital Alliance Global Group), a global cybersecurity and digital transformation platform. He writes monthly on cyber governance for boards navigating the AI era. If this edition was useful, subscribe to Cyber Governance in the AI Era and follow him on LinkedIn for the next one. Protect and Transform.
