Pythagorin @ili_kala

@NBCNews The DOJ's arrest announcement opens by calling OpenAI a "super intelligence (SI), also known as artificial intelligence, company." We haven't demonstrated superintelligence. We've already given it an official definition and a counterintelligence priority.

@CBSNews The DOJ's arrest announcement opens by calling OpenAI a "super intelligence (SI), also known as artificial intelligence, company." We haven't demonstrated superintelligence. We've already given it an official definition and a counterintelligence priority.

The Justice Department's arrest announcement introduces OpenAI as a "super intelligence (SI), also known as artificial intelligence, company." We haven't demonstrated superintelligence. We've already given it an official definition and a counterintelligence priority.

Reuters' October 13 investigation into the negotiations over publication rights and editorial control in independent AI safety evaluations. reuters.com/world/behind-a…

OpenAI and Anthropic are reportedly negotiating what their outside safety evaluators may publish. The evaluators can't block a release. Publication is their strongest public leverage. We've spent months asking who holds the pen. The new question is who holds the black marker.

@NBCNews In March a Texas judge dismissed X Corp's claim that advertisers illegally boycotted it. Now DOJ is investigating a group boycott by networks that refused to fill a banned rival's pool slot. The refusal this time is solidarity with a competitor the government shut out.

October 2020: a House antitrust report finds Big Tech gatekeeping makes the internet less free. Privacy activists celebrate. October 2026: in four countries, major Android stores require developers verified by Google. Antitrust has found a new cartel: newsrooms.

@ABC When frontier labs coordinate to slow down for safety, who gave private competitors the authority to set the pace? If they can’t coordinate, who stops competition from setting it for all of us?

@AP When frontier labs coordinate to slow down for safety, who gave private competitors the authority to set the pace? If they can’t coordinate, who stops competition from setting it for all of us?

@Independent The real threshold comes before full recursive self-improvement. Once AI does most of the work while humans still formally supervise it, "human in the loop" tells us very little. What can that human independently see, judge, and still change?

Four frontier labs, four "AI breakouts." Now it looks like one common cause at one testing vendor. But none were caught as they happened. Anthropic had to scan 481M transcripts to find its incidents after the fact. Monitoring is not the same as recognition. thenextweb.com/news/irregular…

Silent delegation: you still sign off, but the agent already chose the evidence, ranked the options, and framed the trade-offs. Nothing was delegated. Part of the judgment moved anyway. Authority maps show who signs. Not what shaped the decision before it reached them. prnewswire.com/news-releases/…

Opus refused a remote-instance exploit. Hacktron routed its own Discourse box through rce[.]ee/ctf-forum to look like a CTF target. Then it complied. The refusal depended on the model's read of context. That context was attacker-supplied. Three researchers carried it across.

Hacktron AI @HacktronAI

a month ago

On July 25, our team hacked OpenAI. It took us less than 72 hours. Two vulnerabilities chained together gave us access to ChatGPT and Codex accounts belonging to OpenAI employees. We demonstrated the impact with a harmless PR in OpenAI’s internal monorepo. The full chain:

84 391 3K 403K 1K

@HacktronAI “Claude hacked OpenAI” works at headline resolution. Claude compressed exploit-development time; SSO, federated identity and connected tools set the blast radius. Capability found the path. Architecture decided what it could reach.

@CatholicNewsSvc "Conscience" is the key word here. Keeping those decisions in human hands protects the place where conscience acts. It does not form the conscience itself. What forms the person who remains in charge?

Leo’s use of “conscience” goes further than a standard human-in-the-loop argument. Keeping the final decision human does not guarantee good judgment. AI can preserve formal human authority while still amplifying fear, ego, habit or unexamined direction. The harder problem is what governs the human who remains in charge.

Leo's words: a risk of becoming accustomed to "delegating to algorithms decisions that are the sole preserve of the human conscience." Vatican transcript: vatican.va/content/leo-xi…

Pope Leo XIV warned against delegating to algorithms decisions that are "the sole preserve of the human conscience." But keeping a decision human is not enough. AI amplifies the direction we already have. So what governs the conscience holding the tool?

Brilliant 800m battle between Werro and Hodgkinson. But how do you analyse every tactical detail and completely ignore the obvious biomechanical factor? 1.82m vs 1.70m means very different stride mechanics. Hodgkinson getting that close makes the comparison even more striking. @ert_sports #Birmingham2026 #EuropeanAthletics