Threats that are new or growing fast, each backed by dated signals from vendors and agencies. "Emerging" means new or early; "Rising" means established but growing. The advice is MGB's own.
Rising/2 signals · since 2024
ClickFix and fake CAPTCHA
What it is
A fake CAPTCHA or error page tricks the person into pasting and running the attacker's command themselves, which sidesteps controls built for files and links.
Why it matters
CrowdStrike counted a 563% rise in fake CAPTCHA incidents in 2025. It needs no exploit, so it works against fully patched estates.
What to do
Restrict the Run dialog and PowerShell for standard users, alert on commands launched from pasted input, and tell staff that no genuine CAPTCHA asks them to paste anything.
Signals
18 Nov 2026 MGB's explainer on how the paste-and-run attack works, why it slips past controls and how to defend. MGB post →
24 Feb 2026 CrowdStrike's 2026 Global Threat Report reports fake CAPTCHA incidents up 563% in 2025. CrowdStrike ↗
Rising/2 signals · since 2025
Vendor scripts and OAuth integrations as the way in
What it is
Attackers compromise a trusted vendor or integration instead of the target, so the malicious code or token arrives from a source the victim already trusts.
Why it matters
A flaw in a website-experimentation platform let an attacker inject script from the vendor's own CDN into at least 178 sites. In 2025 stolen OAuth tokens from a sales tool were used to pull data from Salesforce instances.
What to do
Treat vendor scripts as production code and keep them off checkout pages. Inventory third-party integrations and OAuth tokens, and revoke the ones you do not need.
Signals
17 Nov 2026 Unit 42 details a platform flaw that let an attacker serve script from the vendor's CDN, hitting at least 178 sites. Unit 42 ↗
26 Aug 2025 Google Threat Intelligence reports stolen OAuth tokens from the Salesloft Drift integration used to pull data from Salesforce instances. Google Cloud ↗
Emerging/1 signal · since 2026
Service-outage phishing lures
What it is
Phishing that poses as a service notice, claiming held mail or access will be lost unless the person acts, to steal credentials.
Why it matters
Microsoft reports a high-volume campaign using Microsoft 365 outage lures across several industry sectors. The urgency and borrowed trust make it convincing.
What to do
Use phishing-resistant MFA and Safe Links, and tell staff to check the admin centre's service health rather than a link in an email.
Signals
16 Nov 2026 Microsoft describes phishing that posed as Microsoft 365 service notices, claiming held mail would be deleted within 24 hours. Microsoft ↗
Emerging/4 signals · since 2026
AI-driven, machine-speed intrusions
What it is
Attackers are using agentic AI frameworks to run many intrusions in parallel and at machine speed, alongside conventional tools.
Why it matters
CrowdStrike reports an 89% rise in attacks using AI in 2025, and the NCSC judges AI-powered malware highly likely to feature in a ransomware attack on UK critical infrastructure by 2028.
What to do
Plan for faster, parallel attacks: phishing-resistant MFA, risk-based Conditional Access, rotating exposed signing keys, patching internet-facing appliances and detection tuned for machine-speed activity.
Signals
6 Nov 2026 Microsoft tracks a group pairing an agentic AI offensive framework with conventional tools against financial organisations. Microsoft ↗
3 Nov 2026 The NCSC's State of the Cyber Nation judges AI-powered malware highly likely in a ransomware attack on UK critical infrastructure by 2028. NCSC ↗
29 Oct 2026 A China-nexus group used an agentic AI framework to run parallel intrusions against more than 200 organisations. CrowdStrike ↗
8 Oct 2026 The NCSC and eight partners link a China-based company to AI-assisted network compromise and data theft worldwide. NCSC ↗
Emerging/2 signals · since 2026
Attacks on AI agents and prompt injection
What it is
Attackers target the AI agents organisations deploy, including how they plan and break goals into steps, not only the prompt.
Why it matters
CrowdStrike's prompt injection taxonomy has grown to 222 techniques, and Microsoft's Digital Defense Report 2026 lists attacks on AI as a surface in its own right.
What to do
Secure the agent's plan as well as the prompt, limit what each agent can access and do, and keep a record of what each agent did.
Signals
29 Oct 2026 CrowdStrike's taxonomy grows to 222 prompt injection techniques, adding attacks on how agents plan. CrowdStrike ↗
Oct 2026 Microsoft's Digital Defense Report 2026 names attacks on AI as a surface among its key takeaways. Microsoft ↗
Rising/2 signals · since 2025
North Korea-linked remote-worker schemes
What it is
Operatives posing as remote workers get hired and use laptop farms in the victim's country to look local.
Why it matters
CrowdStrike reports North Korea-nexus activity up 130% in 2025, and the FBI disrupted several US laptop farms with CrowdStrike's help in October 2026.
What to do
Verify identity in hiring, flag remote-access tools and look for location mismatches rather than relying on static indicator lists.
Signals
13 Oct 2026 The FBI and CrowdStrike disrupted several US laptop farms used by North Korea-linked operatives posing as remote workers. CrowdStrike ↗
24 Feb 2026 CrowdStrike's 2026 Global Threat Report reports North Korea-nexus activity up 130% in 2025. CrowdStrike ↗
Rising/2 signals · since 2025
Trusted-contact and help-desk impersonation
What it is
Attackers persuade a person to hand over access by posing as someone trusted, such as a colleague or a contact, instead of breaking in technically.
Why it matters
ASOS says a login was obtained this way in 2026, and the NCSC said attackers may have posed as staff to IT help desks in the 2025 retail incidents.
What to do
Verify identity out of band before any privileged reset, require phishing-resistant MFA and review help-desk procedures for accounts with elevated access.
Signals
6 Oct 2026 MGB's incident file on ASOS: an employee login obtained by impersonating a trusted contact was used to send an unauthorised push notification. MGB incident file →
4 May 2025 The NCSC notes attackers may have posed as staff to IT help desks to get password and MFA resets. NCSC ↗