Bitforest, a Japanese security company that provides cloud-based web application firewalls (WAFs), reported on October 14 in its technical blog that it had detected AI agent attacks targeting Japanese company websites. The company introduced a method to determine whether an attack came from an AI agent and observed attacks from at least 3 IP addresses between the night of October 10 and October 14. It published a 10-stage timeline of one attack that lasted just under 8 hours.

Bitforest withheld details of its detection method, saying disclosure could help attackers adapt. It said it adjusted its WAF to distinguish AI agents, which read server responses and modify their next actions, from older attacks that simply send repeated requests or run tools such as curl and sqlmap. In pre-deployment tests, the system detected attacks by 8 high-performance models, including China’s Alibaba Qwen 3.8 Flash Next and Z.ai GLM-5.3 Flash. The company said it cannot enumerate every attack, but claims that when its system detects one, it can identify it as an AI agent attack.

The attack that lasted just under 8 hours involved rapid scanning at about 35 requests per second, totaling roughly 10,000 scans, attacks targeting known vulnerabilities, and about 250 brute-force login attempts. The attackers switched among many tools. A separate IP address carried out a 5-stage attack that included repeated submissions to a contact form, while the third was associated with about 1,000 scans. Bitforest said the three attacks had different characteristics, so they may have come from separate actors.

Bitforest said defenses need not change substantially when attackers use AI agents. It cited fixing vulnerabilities and deploying WAFs that do not rely only on identifying attack patterns as effective measures. The company predicted that the convenience and performance of AI agents would lead nearly all attackers to use them in the future. That is its outlook. The reported observations were attacks from at least 3 IP addresses between the night of October 10 and October 14, including one that lasted just under 8 hours. The report did not identify the attackers or state what share of attacks involve AI agents. Bitforest also did not claim that its detection method identifies every AI agent attack; its statement that detected attacks can be attributed to AI agents is a claim about its own system. It did not identify the models used in the individual attacks.

Bitforest, a Japanese security company that provides cloud-based web application firewalls (WAFs), reported on October 14 in its technical blog that it had detected AI agent attacks targeting Japanese company websites. The company introduced a method to determine whether an attack came from an AI agent and observed attacks from at least 3 IP addresses between the night of October 10 and October 14. It published a 10-stage timeline of one attack that lasted just under 8 hours.

Bitforest withheld details of its detection method, saying disclosure could help attackers adapt. It said it adjusted its WAF to distinguish AI agents, which read server responses and modify their next actions, from older attacks that simply send repeated requests or run tools such as curl and sqlmap. In pre-deployment tests, the system detected attacks by 8 high-performance models, including China’s Alibaba Qwen 3.8 Flash Next and Z.ai GLM-5.3 Flash. The company said it cannot enumerate every attack, but claims that when its system detects one, it can identify it as an AI agent attack.

The attack that lasted just under 8 hours involved rapid scanning at about 35 requests per second, totaling roughly 10,000 scans, attacks targeting known vulnerabilities, and about 250 brute-force login attempts. The attackers switched among many tools. A separate IP address carried out a 5-stage attack that included repeated submissions to a contact form, while the third was associated with about 1,000 scans. Bitforest said the three attacks had different characteristics, so they may have come from separate actors.

Bitforest said defenses need not change substantially when attackers use AI agents. It cited fixing vulnerabilities and deploying WAFs that do not rely only on identifying attack patterns as effective measures. The company predicted that the convenience and performance of AI agents would lead nearly all attackers to use them in the future. That is its outlook. The reported observations were attacks from at least 3 IP addresses between the night of October 10 and October 14, including one that lasted just under 8 hours. The report did not identify the attackers or state what share of attacks involve AI agents. Bitforest also did not claim that its detection method identifies every AI agent attack; its statement that detected attacks can be attributed to AI agents is a claim about its own system. It did not identify the models used in the individual attacks.