Home Today’s Briefing Deep Dives In Play Trending For You Discover My Stack Saved Off Radar Search

Loading today’s stories

Today

AI-assisted reporting, grounded in a per-article claim ledger. Perspective Coverage and Reality Strip separate source viewpoints from evidence, adoption, hype, incentives, and confidence.

Today

Front Page
Today’s Briefing
Newsletters
Trending
For You
Discover
My Stack
Saved
Off Radar
Search

Sections

Security
Build
Invest
Product
Leadership
Science

Company

About
Mission
How this is made
Evidence Index
The desks
The sources we read
Corrections
Press and brand
The daily edition

Trending now

Open-Weight Model Publishing
Agentic Coding Tools
Frontier Model Cyber Capability
Software Supply Chain Security
Ransomware-as-a-Service
Vulnerability Disclosure
Home Endpoints as Corporate Risk
Self-Hosted Media Server Security
Autonomous cargo aviation
Humanoid Robotics

How this is made

Today For You My Stack Off Radar Search

Topic

Dual-Use AI Safeguards

Refusal policies, jailbreak hardening and misalignment containment applied to models with exploit-development ability, and the timelines on which they are relaxed.

Current clusters

security1 distinct publisher

OpenAI gates a 100% ExploitBench model behind refusals it plans to loosen in weeks

Astra scored a perfect 100% on OpenAI's own exploit-development benchmark, against 78.5% for GPT-5.6 Sol, and the shipped model's refusal to write proof-of-concept code is a policy the company has already said it will relax.

Publishers:thehackernews.com

Reality

Evidence27
Adoption21
Hype gap+46
Incentives84

Confidence58

Dual-Use AI Safeguards — Clarity Today