[choldstare]

we have to treat these vulnerabilities basically as phishing

[lacunary]

so, train the llms by sending them fake prompt injection attempts once a month and then requiring them to perform remedial security training if they fall for it?