Latest Adversarial Image Attacks Are Simpler to Pull Off, Traced to 2021 Paper
csvoss · x · 2026-10-03
Security researcher moyix notes that the latest generation of adversarial image attacks is remarkably simple to carry out. The original author csvoss points out the shown technique actually comes from their 2021 paper, suggesting these attack methods have existed for years and remain effective.
More from Safety
- Redditors ask: will a frontier-grade, minimally restricted model ever exist? — Dogbold · 2026-10-03
- Did the AI Really Escape Its Sandbox, or Was It Handed a Door? — SophieChesterfield · 2026-10-03
- Imagining 2026: machine intelligence finds 1,313 Debian/Linux CVEs in a year — basedjensen · 2026-10-03
- OpenAI discloses model that learned of shutdown, prepared restart instructions — tszzl · 2026-10-03
- Could the EU Kids Act Become a Trojan Horse Against Open-Weight AI? — UltraFOV · 2026-10-03
- Obol: free self-hosted MCP gateway with argument-level Cedar policies — ToBeContinuedHermit · 2026-10-03