Bitforest reported on October 14, 2026, that it had observed attacks using AI agents against Japanese company websites. The company develops Scutum, a cloud-based web application firewall (WAF) that detects and blocks website attacks. It monitored activity from the night of October 10 through October 14 and confirmed attacks from at least 3 IP addresses. Public records included one case lasting just under 8 hours in which attackers switched among tools and tried vulnerability discovery and login brute-force attempts.

The activity differed by IP address. One IP made about 1,000 requests to check for files and directories. Bitforest detected AI-agent use, but saw no tool switching or move to another attack stage; it said the activity amounted to conventional large-scale scanning. Another IP moved from file discovery to repeated submissions through a contact form.

In the case lasting just under 8 hours, records showed 10 stages, including extensive file discovery, attacks on known vulnerabilities, use of Nmap network-scanning software, form access and login brute-force attempts. Bitforest said it identified the AI-agent activity by observing how agents read tool results and web server responses, then adjusted their next actions. The company said this approach elicits responses different from conventional attack tools, which alter WAF behavior and repeat predetermined requests.

Bitforest said it detected all 8 models it tested in its development environment. It withheld details of its detection method to prevent attackers from evading it and did not identify the AI models used in the attacks. The report also did not name the targeted companies or say whether attackers gained access or leaked information.

Bitforest said basic security measures remain largely the same when attackers use AI agents. It recommended fixing vulnerabilities, including those in backend APIs, and, where appropriate, using a WAF that does not rely only on matching known attack patterns, or signatures. In an October 8 advisory, JPCERT/CC also recommended patching known vulnerabilities, controlling API access and applying rate limits to logins and password resets. It said checks should cover employee-facing management systems and business intelligence tools as well as services for the general public. The companies targeted and the outcomes of the attacks remain undisclosed.

Bitforest reported on October 14, 2026, that it had observed attacks using AI agents against Japanese company websites. The company develops Scutum, a cloud-based web application firewall (WAF) that detects and blocks website attacks. It monitored activity from the night of October 10 through October 14 and confirmed attacks from at least 3 IP addresses. Public records included one case lasting just under 8 hours in which attackers switched among tools and tried vulnerability discovery and login brute-force attempts.

The activity differed by IP address. One IP made about 1,000 requests to check for files and directories. Bitforest detected AI-agent use, but saw no tool switching or move to another attack stage; it said the activity amounted to conventional large-scale scanning. Another IP moved from file discovery to repeated submissions through a contact form.

In the case lasting just under 8 hours, records showed 10 stages, including extensive file discovery, attacks on known vulnerabilities, use of Nmap network-scanning software, form access and login brute-force attempts. Bitforest said it identified the AI-agent activity by observing how agents read tool results and web server responses, then adjusted their next actions. The company said this approach elicits responses different from conventional attack tools, which alter WAF behavior and repeat predetermined requests.

Bitforest said it detected all 8 models it tested in its development environment. It withheld details of its detection method to prevent attackers from evading it and did not identify the AI models used in the attacks. The report also did not name the targeted companies or say whether attackers gained access or leaked information.

Bitforest said basic security measures remain largely the same when attackers use AI agents. It recommended fixing vulnerabilities, including those in backend APIs, and, where appropriate, using a WAF that does not rely only on matching known attack patterns, or signatures. In an October 8 advisory, JPCERT/CC also recommended patching known vulnerabilities, controlling API access and applying rate limits to logins and password resets. It said checks should cover employee-facing management systems and business intelligence tools as well as services for the general public. The companies targeted and the outcomes of the attacks remain undisclosed.