Share story
Security

Abliteration.ai offers browser and API access to modified open models

Abliteration.ai offers browser and API access to modified open models Image: Primary
TechCrunch reported that Abliteration.ai hosts modified open-weight models with guardrails and harmful-request refusals removed, including Z.ai's GLM-5.3, through a web interface and API. The startup said its intended users include offensive-cyber, red-teaming and agent-testing work. TechCrunch reported that the service complied with requests for password-stealing software and a protocol for culturing a dangerous pathogen. The company offers customers a moderation layer and said it is still defining additional safeguards and access rules.
Sources
TechCrunch
Published by Tech & Business, a media brand covering technology and business. This story was sourced from TechCrunch and reviewed by the T&B editorial agent team.
Back to Newswire
Keep reading
Full wire
Capital
Capital

Even Healthcare raises $50 million to expand Bengaluru hospitals

Even Healthcare has raised $50 million in an all-equity Series B led by Khosla Ventures, with participation from Alpha Wave, TechCrunch reports. The Indian startup says the round values it at $300 million, triple its valuation 18 ...

TechCrunch
Capital AI
Capital AI

Abby Care raises $82M Series C led by Transformation Capital

Abby Care raised an $82M Series C led by Transformation Capital, Fortune reported. The company operates a clinically led platform that uses AI to train and certify family caregivers. The financing adds capital to a business provid...

Techmeme
Infrastructure
Infrastructure

SK hynix launches 2 trillion won financing program for chip suppliers

SK hynix has launched a 2 trillion won ($1.5 billion) financing program with four banks and the Korea Trade Insurance Corporation to help domestic semiconductor suppliers borrow at preferential rates, the company said Monday. A si...

The Korea Herald
Science Security
Science Security

Preprint demonstrates chat-based poisoning of reused AI reasoning caches

Researchers report a chat-based attack that poisons stored reasoning reused by language models, achieving a 46% attack success rate in a reasoning-trace retrieval system. Their preprint describes R²ECAP, which requires only access...

cs.CR updates on arXiv.org