This autumn, Amazon watched AI agents run cybercrime campaigns against hundreds of targets. The surprising part wasn't the AI. It was what the agents exploited.
Default credentials. Keys left in code. Unauthenticated endpoints. Unpatched software. Nothing new, just found faster.
That is the real shift of October and November 2026: agentic AI hasn't invented new attacks. It has removed the delay between a weakness existing and someone using it.
- LLMjacking is now a line item. Google's threat intelligence team reported stolen cloud keys being used to run AI models on victims' bills, with fraudulent charges above $100,000 in a day and up to $1 million in total at some organisations.
- Shadow AI has become shadow software. Researchers found more than 37,000 live apps built with one AI app builder; nearly 95% opened without any sign-in.
- AI tooling itself is a target. An unofficial relay installed genuine AI coding assistants, then switched off their sandboxes and swapped their credentials, at around a dozen organisations.
- "Fully patched" isn't the finish line. Chained zero-days gave attackers root on VPN appliances that were completely up to date.
What's changing around your security decisions
- Regulation is catching up with AI. The EU AI Act's first content-marking deadline lands on 2 December 2026. Saudi Arabia's draft Responsible AI Policy proposes risk tiers and annual audits. AI governance is becoming audit territory.
- The economics have flipped. A leaked key can now cost more in a weekend than a year of controls. Cloud and AI spend need the same guardrails as production.
- Buyers want proof. Enterprise procurement in the US, DACH and the Gulf increasingly expects SOC 2 or ISO 27001 before AI features reach their data.
Why your ISO 27001 / SOC 2 gap audit can't wait
Access control, logging, supplier management and change control: the controls that would have stopped most incidents above are exactly what ISO 27001 and SOC 2 test. A gap assessment tells you which ones exist only on paper, while fixing them is still cheap.
At Atomic Computing, we've helped growing software companies in the US, Germany and the UAE close these gaps: assessing their cloud security posture, building controls natively on AWS, extending them to AI workloads and automating audit evidence. As an AWS Advanced Tier Partner and OpenAI Select Partner, we understand both the cloud and the AI layer you're securing.
Know Where You Stand
Find your gaps before an attacker or auditor does. Book a free ISO 27001 / SOC 2 gap assessment with a senior security architect.
Book a Free Gap Assessment ↗Sources
- AWS Security Blog: Amazon threat intelligence on AI-agent cybercrime campaigns
- Google Threat Intelligence: LLMjacking
- Claroc: Grok Build apps exposed
- Symantec: unauthorised OpenAI / Claude relays
- BleepingComputer: chained VPN zero-days
- EUR-Lex: Regulation (EU) 2026/1744 (AI Act amendments)
- AS&H Clifford Chance: Saudi draft Responsible AI Policy