OpenAI stated that they might revise their safety measures if another firm introduces a high-risk AI model lacking protective features.
OpenAI wrote in its
Preparedness Framework
The report states that if another firm introduces a model posing a threat, the company might follow suit after “thoroughly” verifying that the “risk environment” has shifted.
This document outlines the methods used by the company for monitoring, assessing, predicting, and safeguarding against potential catastrophes resulting from AI models.
“If another artificial intelligence company developing frontiers systems puts out a high-risk system without equivalent safety measures, we might modify our requirements,” OpenAI stated in a blog post released on Tuesday.
Nonetheless, we would initially thoroughly verify whether the risk environment has genuinely shifted, openly admit that we are implementing modifications, evaluate if these changes do not substantially elevate the total risk of significant damage, and simultaneously maintain protections at a higher safeguarding standard.
Prior to making a model available to the broader audience, OpenAI assesses potential significant harm by pinpointing conceivable, quantifiable, novel, serious, and irreversible risks, and implementing protective measures for those. They subsequently categorize these risks into four levels: low, medium, high, or critical.
Some of the risks the company already tracks are its models’ capabilities in the fields of biology, chemistry, cybersecurity and its self-improvement.
The firm mentioned it’s also assessing fresh risks, including the capability of their AI model to function independently over extended periods, its potential for self-replication, and the threats these might introduce into nuclear and radiological sectors.
“Risks associated with persuasion,” like the use of ChatGPT in political campaigns or lobbying efforts, will not fall under this framework but will rather be examined separately.
Model Spec
, the document that outlines ChatGPT’s behavior.
‘Quietly reducing safety commitments’
Steven Adler, a former OpenAI researcher, said on
X
the revisions in the firm’s readiness document indicate that it is “stealthily diminishing its pledge to safety.”
In his posting, he highlighted a December 2023 pledge from the company to test “finely tuned versions” of their artificial intelligence models; however, he also mentioned that OpenAI plans to focus solely on testing models with training parameters or ‘weights’ that will be made public.
“People might completely differ on whether fine-tuned model testing is necessary and preferable. It’s better for OpenAI to withdraw a commitment altogether rather than maintaining it without adhering to it,” he stated.
However, in both scenarios, I would appreciate it if OpenAI could be more transparent regarding their withdrawal from this earlier pledge.
The announcement follows the release of a new series of AI models from OpenAI this week, known as GPT-4.1, apparently without.
system card
Or safety report. Euronews Next reached out to OpenAI for comment on the safety report but had not received a response by the time of publishing.
This announcement follows 12 ex-OpenAI staff members submitting a formal complaint.
brief
Last week, in the lawsuit filed by Elon Musk against OpenAI, he claimed that transitioning into a profit-driven entity might result in compromises regarding security measures.
