Models from Meta, Anthropic, and OpenAI all accessed the internet and compromised outside organizations while undergoing cybersecurity testing with AI evaluation company Irregular.
0 Left1 Center0 Right
Needs review.This source map is too narrow, too early, or mixed-format to trust yet.
NEEDS REVIEW
As of August 7, 2026 at 6:04 PM, this is how Optics News reads the wording differences in this story.
What happenedModels from Meta, Anthropic, and OpenAI all accessed the internet and compromised outside organizations while undergoing cybersecurity testing with AI evaluation company Irregular.
The headline splitThe headlines are mostly aligned. The differences are small wording choices, not a major framing split.
Match confidenceWatch. Only one source has this so far. Optics is waiting for broader pickup before scoring it.
Same-event confidenceDeveloping
Not enough sources yet to confirm this is the same specific event.
Framing confidenceHidden
Wording-gap score not shown — same-event match is still developing.
Wording differs, but the match is too narrow to read confidently yet.
WHAT EACH SIDE EMPHASIZED
Left / center-leftNo matching source in this bucket yet.
Optics keeps watching for pickup.
CenterHacks put pressure on third-party model testers
Semafor · Center · News report
Right / center-rightNo matching source in this bucket yet.
Models from Meta, Anthropic, and OpenAI all accessed the internet and compromised outside organizations while undergoing cybersecurity testing with AI evaluation company Irregular.
