Cybercriminals armed with off-the-shelf generative AI tools compromised more than 600 internet-exposed FortiGate firewalls across 55 countries in just over a month, according to a new incident report from AWS.
The campaign, which ran from mid-January to mid-February, relied less on clever zero-days and more on the equivalent of trying every digital door handle – just at machine speed, with AI lending a hand behind the scenes.
AWS says the financially motivated Russian-speaking crew behind the campaign scanned for exposed FortiGate management interfaces, tried commonly reused or weak credentials, and then hoovered up configuration files once inside, giving them a roadmap of victim networks.
The cloud giant’s security team says the actor used multiple commercial AI tools to generate attack playbooks, scripts, and operational notes, effectively allowing a relatively low-skilled outfit to run a campaign that would previously have required more people or time. Investigators even found evidence of AI-generated code and planning artifacts on compromised infrastructure, suggesting the tools were embedded throughout the workflow rather than just used for the odd bit of scripting.
“The volume and variety of custom tooling would typically indicate a well-resourced development team,” said CJ Moses, CISO at Amazon. “Instead, a single actor or very small group generated this entire toolkit through AI-assisted development.”
Once the firewall was cracked, the attackers pulled configuration files containing administrator and VPN credentials, network topology details, and firewall rules. From there, they moved deeper into environments, going after Active Directory, dumping credentials, and probing for ways to move laterally. Backup systems, including Veeam servers, were also on the shopping list.
AWS says the tooling it observed was functional but rough around the edges, with simplistic parsing logic and the sort of redundant comments that suggest a machine wrote the first draft. That didn’t stop it from being effective enough for broad automation, though the miscreants reportedly tended to abandon targets that put up too much resistance and move on to softer ones, reinforcing the idea that volume rather than finesse was the winning strategy.
Geographically, the activity was opportunistic rather than tightly targeted, with victims spread across multiple regions, including parts of Europe, Asia, Africa, and Latin America. Clusters of activity suggested that some compromises may have enabled access to managed service providers or larger shared environments, amplifying downstream risk.
The report leans heavily on the idea that basic hygiene – keeping management interfaces off the public internet, enforcing multi-factor authentication, and not recycling passwords – would have shut down much of the activity before it got going.
The findings land just weeks after Google warned that criminals are increasingly wiring generative AI directly into their operations, including its own Gemini AI chatbot, for tasks ranging from reconnaissance and target profiling to phishing and malware development. ®
