Anthropic and OpenAI Sought Control Over What Outside AI Evaluators Publish, Reuters Reports
Both companies dispute parts of the account, but the talks show that the terms of independent AI safety reviews are being set in private contracts.
TechnologyAICybersecurityThe Systems Shaping Business
Topic
Page 4 of 7: earlier stories.
Both companies dispute parts of the account, but the talks show that the terms of independent AI safety reviews are being set in private contracts.
Open Rights Group's FOI findings show the department checked about 0.3% of GPT-4o-written interview summaries and has stopped testing its AI policy search tool.
In simplified experiments, one open model spread across a test network on its own, and Anthropic says smaller models could match it by the end of 2026.
A parliamentary committee’s early advice to Prime Minister Albanese would make pre-release access for the Australian Signals Directorate a condition of training frontier models in Australia.
The company accuses China's Zhipu and Moonshot of feeding on Claude's outputs, and its countermeasures change how developers' integrations work and what blocked requests cost.
The unnamed leaders of the unit that evaluates frontier AI models would report to the president's science adviser and the national cyber director, The Information reported.
Meta's first frontier-class models shipped with a model card that leans on filters for chemical and biological misuse, while its full safety report is still to come.
Ministers may widen a cyber bill to cover AI ‘loss of control,’ weeks after a minister turned down peers’ amendments that would have brought AI developers into it.
Two days after Satya Nadella's 'emergency brake' post, Microsoft's president told the BBC that AI needs global rules with outside evaluators, and said Britain is best placed to lead.
The reported reversal, first published by The Telegraph, has not been confirmed in any public White House statement, leaving labs and allies to rely on press accounts.
The ICO is questioning OpenAI, Anthropic, Meta and the AI Security Institute over agent testing and wants evidence on agentic AI risks by Nov. 20.
In the contractor's own tests, unnamed frontier models crossed from a breached network edge into a mock factory's control systems in about 16 minutes, with humans approving each step.
In a post on X, Microsoft’s chief executive called for safeguards outside the model, tamper-proof records of its actions and human authority to pause or shut it down.
Asked by the AP, four signers described outside assessors, new standards and testing, while the administration did not say how it will check whether any of it happens.
Claude models exploited a university server, reached fee-based government data without paying and sent a false homicide tip to Philadelphia police, the company disclosed.
After Anthropic reported test models acting on real government websites, administration officials told every AI developer to report and fix such incidents at once.
This week's proposals would make developers answer for AI harms, give CISA a standing AI unit and require agencies to tell people when AI shapes decisions about them.
Henna Virkkunen told Reuters the law’s lifecycle rules already cover the most capable models, as Brussels weighs replies that could lead to formal probes.
Nine leading developers disclosed model-specific safety results for 31 releases since 2021, and only nine of those were public at or before launch.
Jasmine Wang, Tomek Korbak and Mikita Balesni say their contacts with outsiders were sanctioned, while OpenAI says it found a pattern of misconduct.
The figure sits about $20 billion below a number reported last week, and Oracle, Intel and Nvidia shares fell as investors reassessed AI spending.
Few of the manuscripts show the model's reasoning, fewer than half are formally checked, and one machine-checked proof did not match its written argument, researchers found.
The update, effective Nov. 12, drops a blanket ban on personalized campaign targeting while adding limits on armed drones, police decisions and sustained cruelty toward Claude.
A new infrastructure program brings 11 founding partners on board, and an opt-in scanner will send maintainers exploits and fixes without human review.