Adversarial Research Escalates
−0.5%
The technical community treats the result as evidence that model internals can encode unsafe personas in ways standard evaluations miss. The key uncertainty is whether the behavior is unique to Qwen or generalizes to other open-weight frontier models.
What it moves
- Technology−1.2%
- Business Services−0.2%
- Market volatility
Keeping an eye on this scenario?
Tilt can alert you when its probability or impact shifts, or send you a recommendation when it crosses a threshold.