Security News
The online forum OpenAI employees use for confidential internal communications was breached last year, anonymous sources have told The New York Times.Hackers lifted details about the design of the company's AI technologies from forum posts, but they did not infiltrate the systems where OpenAI actually houses and builds its AI. OpenAI executives announced the incident to the whole company during an all-hands meeting in April 2023, and also informed the board of directors.
Security in brief It's been a week of bad cyber security revelations for OpenAI, after news emerged that the startup failed to report a 2023 breach of its systems to anybody outside the organization, and that its ChatGPT app for macOS was coded without any regard for user privacy. According to an exclusive report from the New York Times, citing a pair of anonymous OpenAI insiders, someone managed to breach a private forum used by OpenAI employees to discuss projects early last year.
About Bruce Schneier I am a public-interest technologist, working at the intersection of security, technology, and people. I've been writing about security issues on my blog since 2004, and in my monthly newsletter since 1998.
With Anthropic's map, the researchers can explore how neuron-like data points, called features, affect a generative AI's output. The researchers go into detail in their paper on scaling and evaluating sparse autoencoders; put very simply, the goal is to make features more understandable - and therefore more steerable - to humans.
OpenAI on Thursday disclosed that it took steps to cut off five covert influence operations (IO) originating from China, Iran, Israel, and Russia that sought to abuse its artificial intelligence...
Your profile can be used to present content that appears more relevant based on your possible interests, such as by adapting the order in which content is shown to you, so that it is even easier for you to find content that matches your interests. Content presented to you on this service can be based on your content personalisation profiles, which can reflect your activity on this or other services, possible interests and personal aspects.
A recent partnership announcement between OpenAI and Stack Overflow has some members concerned that their data is being used without permission and will only lead to inaccurate information being created by LLMs. The partnership was announced on Monday, with OpenAI getting access to Stack Overflow's API and feedback from developers and StackCommerce getting links in ChatGPT back to its source material and receiving assistance using OpenAI models as part of its new OverflowAI platform. Epic Games UI designer Ben said he tried to edit his highest-rated answers and replace them with a message protesting the partnership with OpenAI. "Stack Overflow does not let you delete questions that have accepted answers and many upvotes because it would remove knowledge from the community," Ben posted on Mastodon.
The GPT-4 large language model from OpenAI can exploit real-world vulnerabilities without human intervention, a new study by University of Illinois Urbana-Champaign researchers has found. How successful is GPT-4 at autonomously detecting and exploiting vulnerabilities? GPT-4 can autonomously exploit one-day vulnerabilities.
Your profile can be used to present content that appears more relevant based on your possible interests, such as by adapting the order in which content is shown to you, so that it is even easier for you to find content that matches your interests. Content presented to you on this service can be based on your content personalisation profiles, which can reflect your activity on this or other services, possible interests and personal aspects.
Boffins have managed to pry open closed AI services from OpenAI and Google with an attack that recovers an otherwise hidden portion of transformer models. "We thereby confirm, for the first time, that these black-box models have a hidden dimension of 1024 and 2048, respectively. We also recover the exact hidden dimension size of the gpt-3.5-turbo model, and estimate it would cost under $2,000 in queries to recover the entire projection matrix."