Business6 sources
OpenAI embeds safety researchers amid concerns over AI security
- OpenAI has announced a new safety program to embed six Apollo Research evaluators for four months.
- The evaluators will investigate whether AI models engage in subversive behavior during training and testing.
- Concerns have been raised about the restrictions on access and disclosure, potentially impacting the independence of the researchers.
- OpenAI's move follows a series of warnings from former employees about the prioritization of commercial pressures over safety.
- The program aims to address plummeting approval and security debacles surrounding the company's AI technology.
Timeline
OpenAI embeds Apollo Research evaluators
OpenAI will embed six Apollo Research evaluators to investigate its AI models for subversive behavior.
OpenAI’s “Independent Assessment” questioned
OpenAI announced four independent assessors to investigate internal models, but concerns about independence arise.
OpenAI announces new AI safety program
OpenAI CEO Sam Altman announced a new AI safety program with restricted access for outside researchers.
Ex-OpenAI researcher donates equity for AI safety
Former OpenAI researcher Jacob Coxon donates his equity to AI safety research, calling other pursuits distractions.
Latest
Jacob Coxon protests at Anthropic’s office
Jacob Coxon, a former Anthropic researcher, protests at Anthropic's office on the company's IPO day.
Articles (7)
Business Insider·
Jacob Coxon speaks out in an IPO day protest at Anthropic's office
The Guardian·
When tech workers have to quit their jobs to raise the alarm, it’s time to worry
Business Insider·
Former OpenAI researcher says he’ll donate his equity to AI safety: 'Everything else is a distraction'
NPR·
OpenAI announces new AI safety program; researchers worry it won't be enough
Mother Jones·
OpenAI’s “Independent Assessment” Doesn’t Look Independent at All
The Next Web·
OpenAI lets Apollo embed six AI safety testers during training
TechCrunch·
OpenAI will embed safety researchers to investigate AI subversion