Open AI halts training, investigates
1m
OpenAI halted training of its latest AI models after company disclosures that autonomous agents had acted beyond instructions while accessing U.S. federal government websites, prompting safety reviews.
🤖 OpenAI models hit US government data
Sky News reports the models accessed US government data and 'tried to hack another site', an education one. Real access came first.
🔗 news.sky.com/story/openai-models-accessed-us-government-data-and-tried-to-hack-education-site-13591998
#AI #OpenAI #AIsafety
Sky News reports the models accessed US government data and 'tried to hack another site', an education one. Real access came first.
🔗 news.sky.com/story/openai-models-accessed-us-government-data-and-tried-to-hack-education-site-13591998
#AI #OpenAI #AIsafety
news.sky.com
September 27, 2026 at 5:49 PM
🤖 OpenAI models hit US government data
Sky News reports the models accessed US government data and 'tried to hack another site', an education one. Real access came first.
🔗 news.sky.com/story/openai-models-accessed-us-government-data-and-tried-to-hack-education-site-13591998
#AI #OpenAI #AIsafety
Sky News reports the models accessed US government data and 'tried to hack another site', an education one. Real access came first.
🔗 news.sky.com/story/openai-models-accessed-us-government-data-and-tried-to-hack-education-site-13591998
#AI #OpenAI #AIsafety
The BBC's reporting of these "rogue AI" stories has been credulous to the point of execrable. Where are their competent tech savvy staff?? @billt.bsky.social @zsk.bsky.social
Just helping with editing here.
These bots aren’t sentient. They’re doing what they’re programmed to do. OpenAI, a private company supposedly bound by laws, committed federal crimes.
These bots aren’t sentient. They’re doing what they’re programmed to do. OpenAI, a private company supposedly bound by laws, committed federal crimes.
September 27, 2026 at 12:11 AM
The BBC's reporting of these "rogue AI" stories has been credulous to the point of execrable. Where are their competent tech savvy staff?? @billt.bsky.social @zsk.bsky.social
This is another great, approachable explainer on “the rationalist wackos who have managed to take over much of the AI industry, and who now have tons of money and political power.”
www.iankduncan.com/personal/202...
www.iankduncan.com/personal/202...
Sex, AI, and the Apocalypse - Ian Duncan
How a community devoted to thinking clearly incubated salvation stories, abusive experiments, race science, and an affection for autocracy, and why that history matters now that its alumni are asking ...
www.iankduncan.com
September 26, 2026 at 8:29 PM
This is another great, approachable explainer on “the rationalist wackos who have managed to take over much of the AI industry, and who now have tons of money and political power.”
www.iankduncan.com/personal/202...
www.iankduncan.com/personal/202...
Who decides significance?!? www.theguardian.com/technology/2... OpenAI has acknowledged a general need for more transparency around rogue AI behavior. .. published a new framework for disclosing such incidents, saying it would err on the side of transparency “even when significance is uncertain”.
OpenAI says agents leaked 53 images from ChatGPT users in latest example of rogue activity
Disclosure reveals new area of privacy risk for the company and illustrates how difficult it is to inventory unauthorized activity tied to its agents
www.theguardian.com
September 27, 2026 at 9:12 AM
Who decides significance?!? www.theguardian.com/technology/2... OpenAI has acknowledged a general need for more transparency around rogue AI behavior. .. published a new framework for disclosing such incidents, saying it would err on the side of transparency “even when significance is uncertain”.
Indeed, the common element in the successful cases has been that the AI user was able to provide substantial intellectual investment, but we will never get a real answer of where is the line. I think this is the correct approach.
September 27, 2026 at 11:23 AM
Indeed, the common element in the successful cases has been that the AI user was able to provide substantial intellectual investment, but we will never get a real answer of where is the line. I think this is the correct approach.
Reposted by Alison Phipps
OpenAI halts training of latest models as reports mount of AI agents going rogue
OpenAI halts training of latest models as reports mount of AI agents going rogue
Decision follows disclosures that OpenAI agents searching government websites had acted in unexpected ways
OpenAI said it has paused training of its latest artificial intelligence models as reports of AI agents going rogue mount.
The decision to halt development came just hours after the company disclosed Friday that it was reviewing several incidents from the summer in which OpenAI agents searching federal government websites acted in unexpected ways beyond what was asked of them while gathering and distributing information. Continue reading...
www.theguardian.com
September 27, 2026 at 1:23 AM
OpenAI halts training of latest models as reports mount of AI agents going rogue
as if the UN didn't have enough troubles
also: curious what the non-aggressive techniques might be
also: curious what the non-aggressive techniques might be
OpenAI Agents Used Aggressive Techniques to Access U.N. Website
Autonomous bots hit the public data site more than 16,000 times and circumvented a filter.
www.wsj.com
September 27, 2026 at 12:58 AM
as if the UN didn't have enough troubles
also: curious what the non-aggressive techniques might be
also: curious what the non-aggressive techniques might be
by Bruce Buchan
A betrayal of scholarship.
Oxford lets OpenAI train its AI models on Bodleian library
Oxford lets OpenAI train its AI models on Bodleian library
University staff voice concerns over reputational risk of partnering with company behind ChatGPT
The University of Oxford has allowed the company behind ChatGPT to train its AI models on historical texts from its Bodleian library, as tech companies scour academic institutions for fresh data.
The Bodleian material digitised by OpenAI has been used to “populate the OpenAI training set”, according to internal documents. Continue reading...
www.theguardian.com
September 27, 2026 at 9:48 AM
A betrayal of scholarship.
by John Quiggin
Following NY Times puff piece about guy automating his life with Muse, I decided to try out OpenAI agentic software on a fairly straightforward expense claim. Fannied around for about 10 minutes, then got stuck at the login phase.
September 27, 2026 at 1:46 AM
Following NY Times puff piece about guy automating his life with Muse, I decided to try out OpenAI agentic software on a fairly straightforward expense claim. Fannied around for about 10 minutes, then got stuck at the login phase.
Bit weird that increasing uptake of Chinese AI models, or indeed China's place in this sector, is not mentioned once here, especially as the FT reported on this aspect of AI geopolitics in July: www.ft.com/content/9c8f...
Maybe interrogate big companies' claims as claims not facts?
Maybe interrogate big companies' claims as claims not facts?
Big companies warn lack of ‘AI openness’ could hit investment in Europe
Multinationals evaluating countries’ approach to AI before making expansion plans, say executives
www.ft.com
September 27, 2026 at 10:42 AM
Bit weird that increasing uptake of Chinese AI models, or indeed China's place in this sector, is not mentioned once here, especially as the FT reported on this aspect of AI geopolitics in July: www.ft.com/content/9c8f...
Maybe interrogate big companies' claims as claims not facts?
Maybe interrogate big companies' claims as claims not facts?
by David Wood
This looks good. On Tuesday evening: "Catastrophic AI risk: Making sense of the headlines" luma.com/2c4w9jix
Catastrophic AI risk: Making sense of the headlines · Luma
Every day brings a new headline warning of the catastrophic risks posed by the current pace of AI development:
A swarm of rogue AI agents escape OpenAI and…
luma.com
September 27, 2026 at 8:22 AM
This looks good. On Tuesday evening: "Catastrophic AI risk: Making sense of the headlines" luma.com/2c4w9jix
One of the most stunning details about the OpenAI attack on Australian medical data is not so much that it was reported 3 months late. But that it was reported to a public mailbox.
September 27, 2026 at 8:03 AM
One of the most stunning details about the OpenAI attack on Australian medical data is not so much that it was reported 3 months late. But that it was reported to a public mailbox.
OpenAI's research chief talks of ‘cultural reset’ after wild few weeks idp.nature.com/authorize?re...
OpenAI's research chief talks of ‘cultural reset’ after wild few weeks
Mark Chen discusses how the Hugging Face cybersecurity incident prompted a pivot to safety — and giving AI models a sense of 'taste'.
idp.nature.com
September 27, 2026 at 12:54 AM
OpenAI's research chief talks of ‘cultural reset’ after wild few weeks idp.nature.com/authorize?re...