The news
Nine in 10 open critical and high-severity vulnerabilities at organizations analyzed by Detectify had stayed exposed for more than 90 days, the Swedish application security company said on September 29, 2026. The finding lands as AI tools help attackers probe systems faster.
The figure comes from Detectify's H2 2026 Cyber Hygiene Index, which the company said is based on a sample of 1,300 organizations in the US, UK and Nordic countries. The share of open critical and high-severity flaws older than three months was 97% in the Nordics, 92% in the UK and 86% in the US. Detectify said its assessments test real-world exploitability using a payload-based method, so the backlog consists of verified risks rather than theoretical ones.
The company flagged a second problem it calls shadow AI: AI tools that security teams may not fully see or control. Detectify said it increasingly finds publicly exposed, self-hosted AI platforms and AI-built applications across its customer base, and that organizations with exposed AI tooling tend to resolve critical and high-severity flaws at less than half the rate of its broader customer base.
Rickard Carlsson, Detectify's chief executive and co-founder, said the longer a known issue stays open without an incident, the easier it becomes to treat it as normal. "Exposure can effectively become accepted without anyone ever making a conscious decision," he said. Detectify also noted that a slow fix is not always neglect; a flaw on a low-sensitivity asset or behind other controls may be deliberately deprioritized.
The same day, Cloudflare (NET) published results from pointing frontier AI models at its own web application firewall (WAF), the filter that screens web traffic before it reaches an application. Its tester, which starts from known exploits and rewrites them based on each response, made 1,107 attempts across 45 scenarios against an authorized customer staging environment. Cloudflare said the WAF blocked the vast majority, but human review turned up 49 findings worth investigating, 48 of them in command injection and server-side request forgery, and contributed to three changes to its Managed Ruleset.
The Verge, in a September 28 report, described how smaller hospitals, community banks and nonprofits feel exposed as AI gives attackers more reach. Linda Stevenson, chief operations and information officer at Fisher-Titus Medical Center in Ohio, told the outlet the hospital has one cybersecurity analyst, its first, hired two years ago.
The numbers
- Open critical and high-severity flaws exposed more than 90 days
- 9 in 10 (Detectify sample)
- Share exposed more than three months, by market
- Nordics 97%, UK 92%, US 86%
- Organizations in Detectify's sample
- 1,300
- Cloudflare AI tester attempts on its own WAF
- 1,107 across 45 scenarios
- Findings worth investigating after human review
- 49 (48 in command injection and SSRF)
Why CEOs should care
Boards and CEOs should ask for one number every quarter: how long critical and high-severity flaws on internet-facing systems stay open, split into age bands such as under 30 days, 30 to 90 days and over 90 days. Detectify's point is that risk gets accepted by default when nobody owns the decision. Any flaw left open past an agreed deadline should carry a named owner, a written reason and an expiry date for that exception.
CISOs should start with an inventory of AI tools reachable from the internet, including self-hosted AI platforms and apps built with AI coding tools, since Detectify ties exposed AI tooling to fix rates less than half those of its broader customer base. Cloudflare's test is also a reminder that a firewall is one layer: the company wrote that a payload that slips past a WAF still needs an exploitable application, so keeping software patched remains among the strongest defenses. Ask your WAF provider how it tests rules against AI-generated variations of known attacks.
CFOs and buyers should price the cost of slow fixes. The Verge described how Vivian's Door, an Alabama nonprofit, had its systems pulled offline for three days after suspicious activity, leaving it with a bill of about $3,000. When renewing vendor contracts, ask suppliers for their own remediation deadlines on critical flaws and how they verify a fix actually removed the risk.
The bigger picture
The Verge reported that top AI labs limit access to their most powerful cybersecurity models, such as Anthropic's Mythos and OpenAI's Astra, to a limited list of high-profile organizations, such as Nvidia, Google and Apple, plus infrastructure providers and open-source maintainers. It also recounted that Anthropic said in August 2025 that a cybercrime ring had used its Claude Code tool to extort data from healthcare organizations, emergency services and government entities. Michael Kleinman of the Future of Life Institute told the outlet that the finite number of malicious hackers used to cap the threat, and that is no longer the case.
Defenders are using the same techniques. Cloudflare wrote that language models can mutate attack payloads faster than any human hacker, and the requests that slipped through helped it build new detections for all its customers. What most organizations control is the time between finding a flaw and closing it, and Detectify's data suggests that window often stretches past a full quarter.
What’s next
Detectify recommends continuously discovering new internet-facing assets, giving engineers enough context on exposure and ownership to act on critical findings, and re-testing to confirm fixes work, steps it says can increasingly be automated. Cloudflare said a future post will share white-box tests in which the model knows both the application's weaknesses and the WAF rules guarding it. Expect more vendors to publish AI red-team results, and more boards to ask why known flaws are still open.
What “Fact-checked” means
Fact-checking means testing a story’s facts against the evidence before it is published. This story went through at least two separate checks before this version was published.
- What we checked
- Its names, figures, dates, job titles, quotes and who said what were checked against the story’s sources, including its main source where it could be opened. The headline was checked for accuracy and overstatement.
- How
- A first check reviewed the whole story. If it passed, a second, skeptical check went back to the sources to look for mistakes in the most important facts. If a check flagged the story, it was edited to fix the problems found, and a separate re-check then reviewed the whole story again.
- Who
- The checks are made by our newsroom, as steps kept separate from the writing, under rules set by our editor, Hussein Mukhtar. A story the checks still flag is held for the editor, who decides whether it is fixed, published or dropped.
- If something is wrong
- “Fact-checked” does not mean error-free. If a material error is found after publication, we correct the story and add a note saying what changed. Report an error
Companies in this story
