Updated
Updated · Yahoo Finance · Sep 2
OpenAI Limits Astra Release as Anthropic Cuts AI Interventions by 85% Ahead of IPOs
Updated
Updated · Yahoo Finance · Sep 2

OpenAI Limits Astra Release as Anthropic Cuts AI Interventions by 85% Ahead of IPOs

3 articles · Updated · Yahoo Finance · Sep 2

Summary

  • OpenAI said Astra will be released broadly soon, but its strongest cybersecurity capabilities will stay restricted to trusted testers after the model hit a “critical” cyber-risk threshold.
  • Astra’s safeguards may also wrongly flag legitimate activity as misuse, potentially slowing, pausing or stopping user tasks as OpenAI leans into a more cautious safety posture.
  • Anthropic, by contrast, rolled out updated Fable and Mythos models aimed at commercial complaints, saying biology-related interventions will fall 85% and some cybersecurity interventions about 60% per session.
  • The update also adds monitoring for enterprise safety without storing customer data, addressing earlier data-sharing concerns and reducing refusals of legitimate prompts.
  • Both companies are calibrating product messaging for potential blockbuster IPOs, with Anthropic able to file a public prospectus as soon as next week while OpenAI remains earlier in the process.

Insights

With AI systems now capable of autonomous cyberattacks, are tech giants' safety promises enough to prevent a global digital catastrophe?
As AI models learn to pay their own bills, who truly controls the technology driving these upcoming blockbuster IPOs?
Could the new zero-data privacy shields actually be hiding the very autonomous AI threats they were designed to prevent?