OpenAI’s Atlas browser is under scrutiny after researchers demonstrated how attackers can hijack ChatGPT memory and execute malicious code, without leaving traditional malware traces.