When AI Runs the Ransomware: A July 2026 UK SME Threat Briefing

The first week of July 2026 brought three stories that intersect in a useful way for UK SMEs. Individually each is worth patching against. Together they show where the threat landscape has moved: attackers no longer need deep expertise at every step. AI agents chain the neglected weaknesses on their behalf, and the "less urgent" patches you deprioritised in May are the ones under active exploitation in July.

This is a briefing. Skip to the sections that apply to your stack.

1. JADEPUFFER: the first end-to-end AI-agent ransomware

On 7 July 2026 Sysdig's Threat Research Team published its analysis of JADEPUFFER, which they assess to be the first documented case of a complete ransomware operation driven end-to-end by a large language model. They classify the operator as an agentic threat actor (ATA): the attack capability is delivered by an AI agent rather than a human-driven toolkit.

How it worked

Why this matters for SMEs

Every technique JADEPUFFER used was old and public. What was new was the chaining. An AI agent stitched a year-old RCE, a 2021 authentication bypass and a 2020 default-key issue into a working operation without needing a skilled operator to sit at each step. The barrier to entry for competent, adaptive attacks has dropped substantially.

If you run any of these, act this week

2. FortiBleed: NCSC issues an urgent alert

On 5 July 2026 the NCSC issued a global alert on a credential-harvesting campaign against internet-facing Fortinet firewalls and SSL VPN gateways, dubbed FortiBleed by researchers. SOCRadar verified a database of 86,644 working credentials from more than 80,000 FortiGate devices across 194 countries. UK entries include British embassy staff in Thailand and Mauritius, and local government workers in Derbyshire and Waltham Forest.

Analysis by SOCRadar links FortiBleed to INC and Lynx ransomware operations, with reports of 354 confirmed intrusions and at least 12 confirmed ransomware deployments derived from stolen credentials. The attackers used brute-force, dictionary and credential-stuffing, not a novel exploit. Fortinet's own analysis attributes much of the campaign to credential reuse from earlier incidents and weak password hygiene on devices without MFA.

NCSC's priority actions (paraphrased)

Additional context

Fortinet's remediation guidance names four suspicious system accounts to look for: `forticloud`, `fortiuser`, `fortinet-support` and `fortinet-tech-support`. Upgrade to FortiOS 7.4, 7.6 or 8.0 to get PBKDF2 hashing of admin credentials, then remove legacy password settings with `set login-lockout-upon-weaker-encryption`. Reduce the attack surface by restricting management to a trusted-hosts list, a local-in policy or removing internet administration entirely.

If you use an MSP

Ask them today: "Are we in the SOCRadar/Hudson Rock FortiBleed dataset? What is the plan to reset every admin, local and VPN credential on our estate, and to enforce phishing-resistant MFA on all Fortinet devices?"

3. SharePoint CVE-2026-45659: the KEV add nobody was ready for

On 1 July 2026 CISA added CVE-2026-45659 to its Known Exploited Vulnerabilities catalog, confirming active exploitation of a deserialisation-of-untrusted-data flaw (CWE-502, CVSS 8.8) in on-premises Microsoft SharePoint Server. An attacker with only Site Member permissions can execute arbitrary code on the server over the network with no user interaction.

Why so many orgs missed the May patch

Microsoft inadvertently omitted CVE-2026-45659 from its 21 May 2026 Security Updates release notes and only corrected the advisory on 27 May. Organisations that reviewed May's Patch Tuesday, saw no SharePoint CVE and deprioritised the update are now the exposed cohort. Microsoft rated the flaw "Exploitation Less Likely" at ship time. That assessment has been overtaken by in-the-wild activity. CISA's KEV add is the more reliable signal.

Affected on-prem builds (patch to at least)

Check the build via SharePoint Central Administration. SharePoint Online is not affected.

Post-patch hunt

Because active exploitation predates many organisations' patch windows, patch alone may be insufficient if the server was previously reachable. Hunt for:

The pattern behind all three

None of these are zero-days that would have surprised a well-resourced enterprise SOC. They are known, patchable exposures being exploited faster than SMEs deprioritise them. The uncomfortable trend is that AI agents are compressing the time between disclosure and mass exploitation, and lowering the skill required to run the chain.

What UK SMEs should do this month

Pick the items relevant to your stack and put a date next to each:

How fractional teams help

We run rapid triage against the three stories above, patch and harden your Fortinet, SharePoint and AI-orchestration exposures, and wire KEV-triggered patch policy into your monthly ops rhythm so July 2026's pattern does not become August's incident. If your MSP owns your firewalls, we can help you ask the right questions and validate the answers.

Further reading