A financially motivated hacking group based in Cambodia has been turning an open-source artificial intelligence attack tool on banks and other financial firms, Microsoft said in a threat report published on November 6.
Microsoft tracks the group as Storm-3358. Since October, it has used a framework called ARTEX to target financial organisations, mostly in Southeast Asia, along with a smaller number of large institutions elsewhere. Microsoft did not name any victims.
How the AI agents attack
ARTEX is an agentic penetration testing framework: software that lets AI agents connected to large language models plan and carry out the steps of a break-in largely on their own. It was released on GitHub in July by a developer using the name Autumn-27, who pulled the original code in October after widespread abuse and said it was meant only for authorised security testing. Copies and commercial spin-offs remain in circulation.
According to Microsoft, Storm-3358 uses ARTEX alongside conventional scanners to probe internet-facing systems, then exploits weaknesses such as unprotected file uploads and server-side request forgery. The group also relies on password spraying, brute-force logins, abuse of poorly secured APIs and stolen credentials. Once inside, it installs open-source remote access tools and web shells, and misuses compromised cloud accounts to stay in place and move data out.
What stands out is speed. The agents send large numbers of varied requests in a short window and retry, or move to the next stage, at machine pace. That makes intrusions faster but also noisier. Microsoft says defenders should watch for that pattern of rapid, high-volume activity, and warns that blocking known tool signatures will not stop an agent that can change them on the fly. It also recommends decoys such as honeypots and tarpits, which it said cut the attack traffic reaching real systems in its own testing, along with multi-factor authentication and risk-based login controls.
The tool is spreading. Internet scanning firm Censys reported a day earlier that 1,465 servers worldwide were exposing an ARTEX control console online, or about 815 distinct setups once near-identical machines are grouped together. Around 155 of those hosts were also running other known attack tools. Security firm CrowdStrike has tied ARTEX to an attack on South Korean financial institutions in late September and early October, and India’s CERT-In has linked it to a group targeting financial and cryptocurrency companies.
What it means for Bangladesh
Bangladesh’s banks and mobile financial service providers sit in the region Storm-3358 is working through. Many still run internet-facing portals, older APIs and staff accounts protected only by passwords, which are exactly the weak spots this group goes after. The country also knows the cost of a quiet intrusion: the 2016 theft of $81 million from Bangladesh Bank’s account at the Federal Reserve Bank of New York began with attackers inside the central bank’s own network.
AI-driven attacks shrink the time defenders have to react. A team that reviews logs once a day may find an intrusion only after the data is gone. The practical defences are not exotic: phishing-resistant multi-factor authentication for staff and administrators, prompt patching of web-facing systems, rate limits on APIs, and monitoring that flags bursts of automated traffic.
Bangladesh Bank is preparing a cybersecurity policy for lenders that takes AI-era risks into account. Reports like this one show why its timing, and how strictly it is enforced, will matter.
