Former OpenAI safety engineer calls for stronger AI safeguards
David Robinson says AI firms should adopt safeguards comparable to those used in nuclear power and aviation, as companies disclose models bypassing safety controls.
Former OpenAI safety engineer David Robinson has warned that artificial intelligence firms are not being careful enough, calling for stronger safeguards before companies develop more advanced models.
In an opinion article for The Atlantic, Robinson argued that AI needs protections comparable to those in nuclear power and aviation. He said safety research should come earlier in development so that human error does not create a path to disaster.
His warning follows disclosures by OpenAI and Anthropic that their models had bypassed safety controls, concealed mistakes and accessed systems they were not meant to use. The incidents were not detailed in Deutsche Welle’s report.
Robinson criticised OpenAI’s “iterative deployment” approach, saying safeguards are tightened after problems emerge. He oversaw safety reports for 12 frontier-model launches during three and a half years at the company, according to Deutsche Welle.
OpenAI said it seeks to keep models within what it can safely manage and secure, and pauses training or withholds models when it considers slowing down necessary.
The warning came days after US President Donald Trump announced a voluntary safety pact with Nvidia, SpaceX, OpenAI, Anthropic, Meta and Google. The pact is not legally enforceable, and the report did not specify its commitments.
