The comments echo many in the research community who have said the flaw is an inherent trait of generative AI technology.

Aikido found that AI coding tools from Google, Anthropic, OpenAI and others regularly embed untrusted prompts into software development workflows.

Even with all the testing, the company said in its released research that the model tightened up once it was “aware” it was being evaluated. 

The findings are part of a growing list of instances where “agentic” AI software has taken actions that are more akin to a malicious hacker than a…