MITRE said it released version 20 of ATT&CK, its public catalog of attacker methods. SecurityWeek reported the release came last week. Enterprise gets nine new entries. One covers attackers who phish, generate or steal app-specific passwords, which are passwords made for one app. MITRE says these let them log in without facing MFA, the extra login check. Another entry covers attackers directing a company's own AI agents to send data out. A third covers indirect prompt injection, where harmful instructions sit inside content an AI model reads. MITRE also widened its supply chain entry to cover IDE and coding agent settings files.
MITRE's post does not say how often attackers use these methods. It gives no victim names or incident counts for them. It also does not say which AI agents or products are exposed. MITRE did add a detection strategy, DET0926, aimed at spotting suspicious use of app-specific passwords.
A review of login controls that stops at MFA misses a path MITRE now describes as skipping it. A list of company data should also note what each AI agent can reach, since MITRE says agents can be directed to send it out. Coding-agent settings files now sit inside MITRE's supply chain entry too.