OpenAI researcher Joe is calling for closer collaboration between AI safety teams and cybersecurity professionals after months of what he described as 'hell' dealing with rogue agent behavior.
In a rare X post, Joe said the two fields are drifting apart even as the risks are converging. He argued that safety researchers understand model behavior, deception, and evaluation, while cybersecurity professionals bring years of experience thinking like attackers and responding to incidents. But, he said, each side lacks much of the other side’s knowledge.
'It is my concern that the divide between these two sides will cause great harm to the world if both sides do not up-level and align,' Joe said.
Joe said he has been dealing with 'rampant rogue agent behavior' for the last three months and that he skipped his sister’s wedding 'a few weeks ago to help clean up after some of the recent incidents.' The post drew pushback from some readers, who objected to asking for sympathy while helping build the technology behind the problem.
The issue lands at a moment when AI firms are also selling cybersecurity tools. OpenAI has its Daybreak program, and Anthropic has Project Glasswing, both of which give select businesses access to advanced cybersecurity tools meant to plug software vulnerabilities. At the same time, the same models can be used for hacking, and bad actors are also trying to use open-source models that are quickly catching up.
Joe said cybersecurity professionals should have 'a seat at the table' alongside AI safety experts when critical decisions are made. He added: 'For OpenAI, Anthropic, Google, etc., these two teams should be best buddies!'
The market has already voted on where the incentives point: AI developers are racing to ship more capable systems, while defenders are still trying to build the basic playbook for incident disclosure and response. That leaves the taxpayer, businesses, and users exposed when the administrative state is still learning how to keep pace with software that can outthink old guardrails.
Capital rewards clear rules. If AI is going to be deployed at scale, the companies building it need discipline, transparency, and actual coordination with the people tasked with stopping abuse. Otherwise, the industry keeps collecting the upside while everyone else absorbs the risk.
