This is why, in my opinion, if AI will bring exponentially more efficiency, it will also bring an exponential increase in security issues. The most common vuln

This is why, in my opinion, if AI will bring exponentially more efficiency, it will also bring an exponential increase in security issues. The most common vulnerabilities are due to how input is handled (or not). XSS vulns, SQL injections, other injections, etc. Now, the AI agent itself also produces untrusted input. But it does so, in a way, as an extension of the user using it who is actually a trusted entity. Because of this, it is solely up to the user to validate everything coming out of the AI but most users will not do that thoroughly. What you mention in your comment is a good example of that. If the AI reproduces malicious code e.g. inspired from this campaign, it is essentially an injection attack where the user misses it and didn't properly validate the untrusted input.