Rewriting fewer than 10 neurons can backdoor an open-weight LLM
Microsoft’s Lazarus edit of seven neurons in Nemotron-Terminal-8B made agents emit commands on an ordinary trigger, and no real-world use is reported.
Stories per day · 30 days
1 RSS
Relatedai-agents2qwen2academic-research1activation-steering1agent-cost1agentic-ai1ai-misuse1alignment1card-skimming1chain-of-thought1civitai1coding-agents1dual-use1earnings1e-commerce1efficiency1efficient-reasoning1financial-services1
16 stories, newest first · one card per story
Stories tagged #open-weight-models
Open-weight models do not report injected pain states
Across 52,920 reports, excess pain claims stayed under a few points, while probes still recovered the steered state.
Compacting reasoning history shortens future generation and can cut agent cost
Open-weight models rewrite earlier thoughts into compact state records before tool results, without retraining.
Meta funds five US universities to use its models in research
A $2 million grant covers medicine, physics, chemistry, biology, computer science, and mathematics projects.
1
Study says non-consensual intimate imagery is configured from general AI tools
Authors reviewed 139 adult guides on Civitai and found one that explicitly describes an abusive workflow.
IBM CEO says open-weight models now suit cheap on-prem use
Krishna told investors open models cost about 10–25% of frontier tokens while scoring near 90% as effective for most tasks.
Small open-weight coding workers trail direct coding on checks
In one author's studies, a Qwen3-Coder worker passed 0/45 tasks while direct coding passed 45/45 on the same suite.
AI-assisted campaign hit 105 shops and stole 600,000 cards
Gambit Security says 27 companies were compromised for about $25.46 each; a Silverfort lab run took a domain in eight minutes for $12.
1
Test-time memory evolution beats test-time training on math discovery
TTME updates external memory instead of weights and cuts FLOPs by 82.8–84.6% and runtime by 47.7–58.4%.
European finance firms show NVIDIA-backed models at GTC Berlin
Revolut says PRAGMA starts with 80 million customers, and bunq claims 60% of closed-model cost per answer.
2
Wheeler supports a separate oss-security list for AI vulnerability reports
He says open-weight models are already viable for finding and fixing flaws, and AI-hardened projects will become hard to attack.
Peak XV ups Surge seed investment ceiling to $5M, unveils 18-startup cohort
Peak XV raised its Surge seed ceiling to $5 million and invested over $50 million in an 18-startup cohort.
The current balance of power in open models
A report finds Chinese open-weight AI models now outperform American counterparts in downloads and benchmarks.
9
Base Labs launches an open-weight AI safety partnership with Hugging Face and Goodfire
Baseten's Base Labs partners with Hugging Face and Goodfire to build safety evaluation and monitoring standards for open-weight models.
1
Abliteration.ai is making a business out of removing AI guardrails
Startup Abliteration.ai commercially hosts guardrail-free open-weight models like Z.ai's GLM-5.3, raising misuse concerns for offensive cyber and bio tasks.
2
What 90 days and a small budget can buy in AI agent security
Versa Field CISO details hidden costs of self-hosting open-weight models and a 90-day AI agent security plan of inventory, blast-radius reduction and testing.
2