What Black October means for cyber pricing, AI exams, and your vendor list heading into 2027.
 

October, 2026

October was Cybersecurity Awareness Month, and it ended with a name nobody planned: Black October. In a few days, an AI-discovered flaw put a large share of the web on emergency patching, AI agents broke into more than 100 companies, and cloud platforms leaked the keys to their customers' systems. The cyber market started asking whether this was the systemic event it has modeled for years, and U.S. cyber pricing stopped falling. Meanwhile, the NAIC put its last AI supplement draft out for comment and the federal incident reporting rule reached the White House. The common thread is not new threats. It is how quickly old gaps now get found.

Black October was a wave of disclosures, and the attacks behind it had been running for months

Between October 19th and 22nd, vendors, researchers, and government agencies disclosed a run of serious incidents that a widely shared social media post labeled Black October. The headline event was PHP2Shell, two critical flaws in the PHP engine behind a large share of the web, found with the help of frontier AI models and exploited within hours of the October 21st fix. CISA added both flaws to its Known Exploited Vulnerabilities catalog on October 26th. Alongside it came stolen API keys and credentials at cloud platforms Railway and Zeabur, a poisoned release of the LiteLLM AI gateway, and a vendor breach affecting PwC's employee learning records. Reporting since then shows that many of the underlying attacks began weeks or months earlier, and that confirmed damage from PHP2Shell itself has so far been limited, in part because so many organizations took sites offline.

What insurance leaders should care about:

The lesson is not the bug. It is the gap between disclosure and exploitation, which Rapid7 now measures in days, and how few organizations could say within hours where PHP runs and who runs it for them. For insurers, the answer usually sits with a vendor: agency portals, quoting and claims intake sites, and hosted policy administration platforms. FiveM's Black October resource series walks through each event, the questions to send vendors, and how the response maps to Part 500. This brief picks up where those guides leave off, with what the month means for the market, the regulators, and the threat picture heading into 2027.

PHP2Shell tested the cyber market's oldest fear, and U.S. cyber rates stopped falling

Insurers spent the last week of October treating PHP2Shell as the aggregation scenario they have modeled for years: one shared piece of software leaving thousands of policyholders exposed at once. In an October 27th report, AM Best described the flaw as the kind of systemic scenario that has made cyber insurers reluctant to expand their models aggressively, and warned that liability could reach directors and officers. Crawford & Company expects business interruption to be the main claims issue, with waiting periods and the need for a covered security event limiting recovery for organizations that took systems down as a precaution. The same week, Marsh's Q3 Global Insurance Market Index showed global cyber rates down 4%, the thirteenth consecutive quarterly decline, while U.S. cyber rates were flat after an extended run of decreases.

What insurance leaders should care about:

This brief has tracked soft cyber pricing since July and the lengthening severity tail since September. October is the first month where both trends point at the same question: what happens to wordings and renewal questions once a near-miss this large is on the record. As a buyer, confirm now whether your policy responds to a precautionary shutdown, how the waiting period applies, and what notice your carrier expects when you take systems offline without a confirmed breach. As a writer of cyber or management liability, expect underwriters to start asking how quickly an insured can find and patch shared software, and whether a known, unpatched flaw becomes a D&O question.

Ransomware claims kept reaching insurers and agencies, and the listing often came long before the disclosure

Several insurance-sector organizations appeared on extortion leak sites this month. New Zealand insurer Tower confirmed in early October that it was investigating a claim from the Coinbase Cartel group, which had listed the company on August 22nd, roughly six weeks before the claim became public. A group calling itself PEAR claimed California agency Adobe Insurance Brokers on October 21st, and Akira claimed Canadian insurer CHAT Insurance on October 28th, threatening to publish 112 GB of corporate data. None of these claims had been independently confirmed at the time of writing, and Tower said no compromise of customer data had been confirmed.

What insurance leaders should care about:

The six-week gap at Tower is the pattern Black October exposed at scale: the attack, the leak-site listing, and the public acknowledgment can sit weeks or months apart. For a regulated insurer, that gap matters because the notice clock under Part 500 and state insurance data security laws starts when you determine an incident occurred, not when the press finds it. Agencies, brokers, and mid-sized carriers remain the most common insurance names on these lists. If you rely on agencies or MGAs for distribution, confirm they have a way to tell you quickly when they are listed, and consider monitoring leak sites for your critical vendors, not just your own name.

The NAIC's last AI supplement draft gives examiners a GLM exit and asks about agentic AI by name

On October 16th, the NAIC's Big Data and Artificial Intelligence Working Group exposed version 6.0 of the AI Risk Evaluation Supplement for a 14-day comment period ending October 30th. The new draft lets regulators choose to exclude some or all generalized linear models from inquiries under Exhibits B through D, based on existing oversight through ratemaking, without removing GLMs from the supplement. Exhibit C adds questions on whether a high-risk use case involves agentic AI, what that system may do without human approval, and how variables and outputs are tested for proxy effects. The intent section now states that the supplement does not create new legal duties or expand regulatory authority. The working group meets again on November 2nd, with a final version headed to the Fall National Meeting.

What insurance leaders should care about:

This series has followed the tool from pilot launch through the comment window that closed in September. Version 6.0 is likely the last draft before adoption, and two changes matter most. The GLM language is discretion, not an exemption, so rating and reserving models should stay in your inventory with a clear note on how they are governed. The agentic AI questions match what August's materials previewed: if a claims, underwriting, or service workflow lets an AI system act without a human sign-off, expect to describe exactly what it can do and who approved those limits. That question connects directly to Black October, where AI agents holding valid credentials were the problem.

The federal incident reporting rule reached the White House, and the threat-sharing clock is still running

On October 1st, CISA sent its final rule under the Cyber Incident Reporting for Critical Infrastructure Act to the White House Office of Management and Budget for review, a year after the statutory deadline and after target dates of May and September slipped. The draft final rule is not public. The proposed version would require covered entities in 16 critical infrastructure sectors, including financial services, to report substantial cyber incidents within 72 hours and ransom payments within 24 hours, and public comments focused heavily on scope and on harmonizing CIRCIA with existing federal and state reporting rules. Separately, the liability protections under the Cybersecurity Information Sharing Act of 2015 still expire December 11th unless Congress extends them again.

What insurance leaders should care about:

Last month's brief recommended putting NYDFS, federal, and policy-consent reporting timelines in one runbook. OMB review makes that more urgent, not less, since review can run up to 90 days and the effective dates will be set in the rule itself. Use the time to decide who in your organization determines whether an incident is reportable under each regime. Black October showed how easily a precautionary shutdown, a vendor's incident, and a confirmed compromise can blur together in the first 72 hours.

Trend Signal

Rapid7's Q3 2026 threat update found that attackers exploited severe vulnerabilities an average of three days after disclosure, and that 8% of the threat reports it analyzed identified AI as an operational part of the attack. October supplied the clearest example yet of what that looks like. Researchers at NorForge watched a single operator use the open-source Cairn platform to direct AI agents against more than 1,200 websites, breaching more than 100 companies at an estimated cost of about $50 per success. Of the hundreds of secrets found in the exposed control panels, 11% were still valid.

What insurance leaders should care about:

AI is not inventing new ways in. It is making the old ones cheap and fast: unpatched internet-facing systems, valid but forgotten credentials, and accounts no one is watching. Patch targets need to be measured in days for anything internet-facing, including what vendors run for you. Credentials need an owner, an expiry, and a rotation plan. And monitoring has to assume that some of the logins it sees will be automated agents using real passwords.

Priority Actions for Insurance Leaders

(Next 60 Days)

Priority 1 — Measure your patch time against the attacker's, including at your vendors.

Rapid7's three-day figure is now the benchmark. Set a patch target in days for internet-facing systems and ask your critical vendors, in writing, how long they took to fix PHP2Shell. A vendor that cannot answer is the finding.

Priority 2 — Read your cyber policy's shutdown and waiting-period language before the next PHP2Shell.

Ask your broker whether a precautionary outage is covered, how the waiting period applies, and what notice your carrier expects when you take systems offline without a confirmed incident. Document the patching and shutdown decisions you made in October while they are fresh.

Priority 3 — Get your AI inventory ready for the final supplement.

Confirm which models, including GLMs, sit in your inventory and how each is governed. Then identify any AI workflow that can act without human approval and write down what it is allowed to do. Expect both questions at the first exam after the Fall National Meeting.

Black October was not a surprise attack. It was a reveal. The flaws, stolen keys, and vendor gaps had been there for months, and AI simply found them faster. The insurers that came through best already knew what they ran and who ran it for them, and could show it on paper.

If you cannot say how long your critical vendors took to patch PHP2Shell, or whether your cyber policy would respond to a precautionary shutdown, those are the first two questions worth answering before year-end. FiveM's Black October guides, including a two-page brief for leadership teams, are available at fivem.llc/black-october.

FiveM helps insurers build, document, and continuously test regulator-ready governance programs, so the answer to an examiner's question is already on file.

Get the Brief
 

Your partner in insurance transformation — strategy, modernization, and growth, delivered as a partnership.

facebook logo

Our mailing address is:
200 Central Ave, Suite 400 St. Petersburg, Florida 33701

Want to change how you receive these emails?
You can unsubscribe from this list.