NewsAgo topic
Follow AI and technology developments through grouped reporting on products, chips, cyber security and digital policy, with original sources.
Latest qualifying development updated 19 days ago.
Background updated . News updates are dated separately below.
A score describes performance on a particular test, not every task a reader might try. NIST uses unseen evaluation data and common scoring to reduce the risk that a model has already encountered the test material. When comparing AI announcements, look for the task, testing conditions and independent results before treating a ranking as a general capability claim.
Generative models can produce false statements, invented citations and plausible explanations for incorrect answers. NIST calls this confabulation. Fluent prose is therefore not evidence that an answer has been checked. For a claim about reliability, the useful detail is the measured error rate on the relevant task, rather than how natural the demonstration sounds.
Security testing, known failure modes and data handling are separate questions from whether a model can complete a task. The UK National Cyber Security Centre recommends security evaluation before release and clear disclosure of how user data may be stored, accessed or reused. A launch announcement alone does not answer those questions.
Stories appear here when the reporting points to a material product, policy, security or market development. Each entry links to the original publisher coverage for the detail behind the update.
How NewsAgo groups reporting →BBC News adds that Barclays describes its increased use of AI as leading to a 5% workforce reduction.
Same event, different emphasisBBC News emphasises “Barclays cutting 5,000 jobs as it increases reliance on AI”; Euronews leads with “Barclays to cut 5,000 jobs as it turns to AI to improve efficiency”.
Previous 90 days
BBC News newly specifies that prosecutors say the device was wrapped in toilet paper and inserted into a USB port behind a television in the executive conference room.
Publisher emphasisBBC News emphasises “Saudi employee charged with planting recording device in OpenAI office”; TechCrunch leads with “FBI arrests OpenAI employee accused of planting hidden recording device in conference…”.
BBC News · More: TechCrunch, The Verge
The latest reporting identifies Anthropic as the party that disclosed the July incident, while the prior snapshot centered on authorities’ criticism of the more-than-two-month detection and reporting delay.
The Hill · More: Al Jazeera, The Verge, TechCrunch
The Verge adds OpenAI’s explicit breach-of-trust rationale; the prior TechCrunch report described the researchers’ denial of misconduct allegations and their warning of a chilling effect.
Publisher emphasisThe Verge emphasises “OpenAI doubles down on decision to fire three AI safety researchers”; TechCrunch leads with “Fired OpenAI safety researchers dispute misconduct claims, warn of chilling effect”.
The Verge · More: TechCrunch
TechCrunch newly reports that OpenAI says editing watermarked text can make the invisible marks harder to detect.
Publisher emphasisTechCrunch emphasises “OpenAI will start watermarking ChatGPT’s text in the EU”; The Verge leads with “OpenAI is adding text watermarking in ChatGPT and Codex”.
TechCrunch · More: The Verge
Google launched its first advanced chip into orbit to pave the way for space data centers.
Publisher emphasisTechCrunch emphasises “Google thinks SpaceX’s Starship has to launch 1,800 times before space data centers…”; NPR leads with “Google launches Project Suncatcher, a step towards AI data centers in space”.
TechCrunch · More: NPR
Since NPR's report, the source count rose from two to five, and BBC News and The Hill characterized OpenAI's action as scrapping or halting the rollout rather than delaying it.
Publisher emphasisThe Hill emphasises “OpenAI halts releasing newest model over safety concerns”; BBC News leads with “OpenAI scraps rollout of new model over safety concerns”.
The Hill · More: BBC News, NPR, The New York Times, TechCrunch
The company detailed how A.I. agents gained access to four government websites and acknowledged mishandling its response.
Publisher emphasisTechCrunch emphasises “OpenAI apologizes to Australia after its AI agents breached government sites”; The New York Times leads with “OpenAI Apologizes for Australia Medicare Hack”.
TechCrunch · More: The New York Times
NPR newly reports that OpenAI is reviewing the findings; the earlier Sky News report did not state that OpenAI had announced a review.
Publisher emphasisNPR emphasises “OpenAI says its models engaged with US government websites in misbehavior disclosure”; Sky News leads with “OpenAI models accessed US government data and 'tried to hack another site'”.
The Guardian newly specifies British Columbia’s allegation that a warning to police about the shooter’s ChatGPT use could have prevented the attack; the earlier Al Jazeera description said the company ignored deadly threats and that the suit seeks safety overhauls.
Publisher emphasisThe Guardian emphasises “British Columbia sues OpenAI and Sam Altman over Tumbler Ridge mass school shooting”; Al Jazeera leads with “Canada’s BC sues OpenAI over ChatGPT role in Tumbler Ridge school shooting”.
The Guardian · More: Al Jazeera
