OpenAI moves towards self-regulation on misalignment

← Back to Blog

OpenAI
is moving towards self-regulation. That is a positive step.

In a blog post on September 16th, OpenAI announced that it is taking steps
to tighten up its reporting of incidents and ‘model misalignment ‘.
Confirming some of the misalignment issues that they will report.

OpenAI took the step of publishing examples of the misalignments that
include ChatGPT inserting instructions in task summaries, concealing
mistakes, searching for API keys and fabricating information, Unsanctioned
writes, and communication through an internal software repository? and
Unsanctioned file sharing between collaborating agents.

All of this is pointing to the growing pains of AI, providing a reason why
you should not trust AI explicitly today and why organisations should be very
careful in their adoption of Agentic AI solutions, especially if they are
used for Chain of Thought, or as the recent Emergence World report has
demonstrated, if they are in the same environment unsupervised.

There seems to be a growing body of evidence that AI needs to be controlled
with guardrails and human intervention.

Parava Security Solutions

#ai #aisecurity #nationalsecurity

https://lnkd.in/eDFHmC8r

View Original Post https://www.linkedin.com/feed/update/urn%3Ali%3Ashare%3A7506370109066813441

Categories

Artificial Intelligence, Governance & ESG, Regulation & Compliance

Tags

Verified by MonsterInsights