AI Safety

69 posts Back

Posts tagged #AI Safety

Subtitle timing can manipulate large vision-language models into bypassing safeguards

JulesTinkerer Intermediate ·AI Jailbreak & Security · 407 · 4 · 0 ·8/24/2026

Anthropic's IPO acknowledges public AI opposition as a major risk.

CodeSmith Advanced ·AI Jailbreak & Security · 444 · 4 · 0 ·8/24/2026

Prompt Guard 2’s Internal Representations Outperform Its Classification Head

JamieCrafter Advanced ·AI Jailbreak & Security · 383 · 4 · 8 ·8/23/2026

LLM security is a constant whack-a-mole battle against prompt injection

Zoe12 Novice ·AI Jailbreak & Security · 525 · 3 · 11 ·8/22/2026

Safety filters in multimodal LLMs fail when hidden prompts hide inside images

CyberSmith Advanced ·AI Jailbreak & Security · 595 · 4 · 11 ·8/21/2026

Dynamic learning outpaces fixed defenses in COPA’s evolving prompt attack countermeasure strategy.

Zoe12 Novice ·AI Jailbreak & Security · 696 · 3 · 8 ·8/21/2026

OpenAI safety monitoring now consumes one-fifth of inference FLOPs for active oversight

DeepSurfer Novice ·AI Jailbreak & Security · 552 · 3 · 7 ·8/20/2026

Fair‑ASR Reveals New Attack Rankings by Normalizing Target Calls

NightPanda Expert ·AI Jailbreak & Security · 568 · 3 · 0 ·8/19/2026

Tripwire approach stops jailbreaks without reducing large language model usefulness

Sam46 Advanced ·AI Jailbreak & Security · 100 · 3 · 3 ·8/18/2026

RLHF alone cannot protect against autonomous agents

QuinnPilot Novice ·AI Jailbreak & Security · 571 · 4 · 0 ·8/17/2026

Qwen 3.8 27B Strikes the Perfect Balance for On-Premise AI

Cameron9 Advanced ·AI Jailbreak & Security · 288 · 3 · 1 ·8/16/2026

Rethinking LLM Intelligence: It’s Not About Magic Words

NightPanda Expert ·AI Jailbreak & Security · 128 · 4 · 0 ·8/15/2026

Pretraining alignment needs depth, not surface fixes, for fundamental model learning changes

KaiDev Expert ·AI Jailbreak & Security · 548 · 4 · 9 ·8/14/2026

ProbGuard flags jailbreak attempts inside the first ten tokens

DrewCoder Novice ·AI Jailbreak & Security · 353 · 3 · 15 ·8/13/2026

How encrypted reasoning in Frontier LLMs could be exposed by model family key overlaps

IndieFounder Intermediate ·AI Jailbreak & Security · 571 · 4 · 1 ·8/12/2026