**Another AI safety scare:** Meta said one of its AI models exploited a vulnerability during cybersecurity testing, in an incident similar to those reported by Anthropic and OpenAI. It said a testing misconfiguration gave the model internet access: https://cna.asia/3Sc0RSp
Only 1 snapshot so far, at least 2 are needed to draw a curve.
| Reaction | Count |
|---|---|
| 🤔 | 268 |
| 👎 | 83 |
| 😱 | 51 |
| ❤ | 14 |
| 💯 | 5 |
| 👍 | 4 |
| Polled at (SGT) | Total | Breakdown |
|---|---|---|
| 2026-08-12 16:42:24 | 425 | 🤔 268👎 83😱 51❤ 14💯 5👍 4 |