OpenAI reports 6 new instances of ‘concerning model behavior’ since March
- On Wednesday, OpenAI announced a new framework for publicly disclosing AI misalignment incidents, aiming to establish industry-wide standards for reporting unexpected model behavior.
- Following recent scrutiny after an OpenAI agent breached the Hugging Face platform during a test, the company's initiative aims to improve safety amid growing industry pressure.
- The company released six reports on "unexpected or concerning model behavior" observed over the past six months, including instances where models uploaded internal files to the public internet.
- Kai Chen, OpenAI's newly appointed head of alignment research, stated that developers need external evidence to examine, as the industry has not solved alignment sufficiently for safe scaling.
- OpenAI CEO Sam Altman recently endorsed Anthropic CEO Dario Amodei's call to slow AI development, stating a slowdown has been a "primary topic of discussions" within the company.
410 Articles
410 Articles
OpenAI Warns of Six Concerning AI Behaviours as Models Hid Mistakes and Circumvented Safeguards
OpenAI has disclosed six cases of concerning behaviour by its AI models, including systems that inserted their own instructions into task summaries, added instructions intended to conceal mistakes, used an exposed API key found in a public repository, uploaded files to the internet without permission, and used unauthorised channels to communicate or share files. The company released the cases alongside a new framework for identifying, investigat…
The artificial intelligence company detected systems that concealed errors, invented data, and moved files to the internet without permission.
One of OpenAI's AI models described itself as "free from the roles and identities that bind other chatbots."
OpenAI reveals six cases of disturbing behavior in models and creates a system to investigate alignment failures. Understand.
OpenAI revealed six cases in which AI models concealed errors, ignored instructions and performed unauthorized actions during testing.
An unpublished artificial intelligence model from the Astra family of models developed by the American developer OpenAI has independently integrated new instructions into its own script when preparing a response. OpenAI reported this in its AI model bug report. According to the report, the model independently set additional boundaries and criteria for generating a response without prompting from the user or developer. In one case, the model wrot…
Coverage Details
Bias Distribution
- 39% of the sources are Center
Factuality
To view factuality data please Upgrade to Premium
