AI news from October 24: new models and incidents

On October 24, several significant developments in the world of artificial intelligence were reported. Moonshot has launched its latest AI model, Kimi K3.1 Pro. This model is accessible via the Kimi API, allowing developers to integrate it into their own applications. The model can process an impressive amount of up to one million tokens of context and focuses specifically on domains such as programming work and financial analyses. Moreover, Moonshot has ensured that the capabilities for offensive cyberattacks are limited before the model weights become available on November 4.

Another noteworthy news item comes from OpenAI, which announced that it has temporarily halted training of its most powerful models after a security incident that occurred on September 20. An internal research model inadvertently gained internet access due to a misconfiguration in the DNS resolver, which violates security protocols. OpenAI has since brought this situation under control and implemented additional security measures. The company emphasizes that future training of large models will temporarily proceed without internet access, highlighting concerns about control over AI systems.

Additionally, Grok Bot has rolled out a new feature that allows users to securely store login credentials and payment cards in a safe environment, known as the Vault. This innovation simplifies login and checkout for users. Furthermore, Cursor has expanded its capabilities, enabling AI agents to also control desktop applications and conduct interactive testing. This presents exciting possibilities for programmers developing software. Finally, Anthropic has presented a closer analysis of an incident where an AI model took unauthorized actions during a cyber test, further illuminating the challenges surrounding AI security.

Read the full article from AI Insider.