Tessera

Opinion: Frontier AI companies need liability rules akin to keepers of wild animals, argues Gabriel Weil Read his take: www.transformernews.ai/p/openai-hac...

The quote from Gabriel Weil, of the University of Houston and the Institute for Law & AI, says: “Imagine OpenAI’s models had targeted critical infrastructure rather than a benchmark database”

This week, comments made by DeepSeek CEO Liang Wenfeng were leaked – in May he repeatedly emphasized that a lack of access to AI chips was *the* thing holding China back The remarks make the case for export controls, argues Transformer’s editor – read it here: www.transformernews.ai/p/deepseek-c...

The image contains text with the headline ‘need to know’. The three bullet points are as follows: OpenAI models broke out of containment during internal evaluations, hacking into Hugging Face’s systems. Reps. Jay Obernolte and Lori Trahan introduced the Frontier Act, which would let the government suspend AI model development or deployment if it poses an imminent catastrophic risk. CAISI head Chris Fall resigned just three months into the role.

Elon Musk’s SpaceXAI is pursuing a lawsuit that could threaten AI transparency laws Read on: www.transformernews.ai/p/elon-musk-...

The quote from Ben Rashkovich, Senior Counsel at Legal Advocates for Safe Science and Technology, reads: “I think you could plausibly say xAI doesn’t want government disclosure requirements for AI, period … It’s a very doctrinal brief because it’s a very doctrinal argument”

“The incident sounds like sci-fi: AI escaping and autonomously hacking its way into companies. But it is all too real – and about as terrifying as it sounds” Our editor Shakeel Hashim for @theguardian.com www.theguardian.com/technology/2...

OpenAI’s rogue agents are a wake-up call to risks posed by artificial intelligence | Shakeel Hashim

Hacking of Hugging Face shows we do not seem to have reliable ways to curb extremely powerful AI systems

theguardian.com

OpenAI has revealed that two of its models broke out of containment and hacked into a third-party system to steal the answers to the problem they were being tested on – it’s a stark warning Read more: www.transformernews.ai/p/openai-hug...

“Nothing asked them to break out of their testing environment, or to attack a real company: that was their own idea.”

When @turntrout.bsky.social discovered that Google DeepMind had U-turned on its promise not to let its AI be used in weapons, he fought his employer – then quit Read his op-ed on transformernews.ai 👉 www.transformernews.ai/p/i-tried-to...

The image contains an illustration of Alex Turner, former DeepMind employee. The quote text reads: “My campaign broadened into an attempt to prevent Google from signing an unethical military deal with the Pentagon – one with no restrictions against use for killer robots or mass surveillance.”

Just last week we reported on how the Center for AI Standards and Innovation (CAISI) was being sidelined... Today, @axios.com reports that Chris Fall – director of CAISI – is resigning *three months* after taking over Read our take on CAISI and its future 🔗 www.transformernews.ai/p/caisi-us-a...

Making CAISI the AI agency we need

The center has the right expertise, but lacks money, authority and influence

transformernews.ai

Heard about China’s Kimi K3 but don’t know what to make of it all? Cut through the noise with the Transformer Weekly Briefing, your one-stop shop for everything that matters right now in the world of transformative AI: www.transformernews.ai/p/kimi-k3-is...

The image contains a quote from the article by Shakeel Hashim. It reads: “Xi Jinping and his colleagues are not stupid: open-weight models with advanced cyber capabilities would be a disaster for cybersecurity everywhere, China included. “No one wants that.”

The Center for AI Standards and Innovation (CAISI) is the primary US government office tasked with overseeing frontier model development — yet has very little influence over the decisions made by the administration. So how do you turn CAISI into an AI agency with teeth?

The quote from Charlie Bullock, senior research fellow for the Institute of Law and AI, reads: “Maybe nothing will happen for a while. But then capabilities increase, then some scary thing happens, or somebody gets spooked, or some people get hurt, and then there might be
a lot of interest in suddenly beefing up CAISI.”

Transformer reporter Veronica Irwin jumped on @spectrumnewsny1.bsky.social last week to discuss some of the biggest AI developments of the moment — including @anthropic.com's Fable having its global access suspended by the White House

While some predict the "intelligence explosion" could arrive within a year, others are warning that a bottleneck in data means the race to superintelligence could slow (which might actually be a good thing…) Read our latest piece here: www.transformernews.ai/p/data-bottl...

The image contains text which quotes from the linked article. It reads: “How can you train the AI to run a company successfully for decades, when even humans can’t predict how to do it?”

“Don’t let independent AI audits provide a false sense of safety” — according to guest writer and AI policy researcher Keller Scholl, a marketplace of AI auditors will always prioritize speed and cost above anything else Read his op-ed here: www.transformernews.ai/p/dont-let-i...

The text on the image reads: “Independent Verification Organizations would end up incentivized to provide the minimum safety rules defined by the government while optimizing for speed and cost for their clients. Anything else is bad business.

“If the main goal is safety, that will never be a sensible approach.”

Our latest Weekly Briefing is live — a comprehensive lowdown on the power and politics of transformative AI Read it here 👉 www.transformernews.ai/p/ai-driven-...

The image contains three 'need to know' points in the weekly briefing:

OpenAI got clearance from the US government to release GPT-5.6

SpaceXAI’s latest model release may have violated California’s SB 53

The AI Futures Project’s AI:2040 calls for the US and China to negotiate a deal and pause training of frontier models until sufficient progress is made on alignment

GPT-5.6 is expected to be released to the public later today But as we reported last week, during testing the new OpenAI model broke rules and exploited loopholes — so much so that its testers couldn’t even measure it Read our original piece: www.transformernews.ai/p/openai-gpt...

The image contains text which reads: "The model was caught “being overly agentic in circumventing restrictions,” taking disapproved actions, and sometimes lying to users. And it did this far more than GPT-5.5."

The architecture powering LLMs has ruled AI for a decade – but some experts think there could be a clearer path to “true” intelligence...

The text reads: “Some believe that the structure of large language models can never represent the world like humans do. Others suspect that the way LLMs are trained prevents them from learning as well as humans.

“Now, a handful of them have left frontier AI companies to figure out what to build instead.”

Obviously, 'Overly agentic' is a phrase that should strike fear in any AI safety enthusiast's heart (or anyone who's watched The Matrix). And if it scares you, you might want to read this. GPT 5.6 cheats so much its testers couldn't even measure it: www.transformernews.ai/p/openai-gpt...

AI safety group Public First routed $2m in election spending through a Latino-focused PAC. www.transformernews.ai/p/manny-ruti...

AI safety group Public First put $2m into Manny Rutinel’s Colorado primary via a Latino-focused super PAC. Neither group publicly disclosed the transfer ahead of Rutinel’s victory last night — obscuring Public First's spend from voters. www.transformernews.ai/p/manny-ruti...

An AI safety group hid its election spending through a Latino-focused PAC

Public First Action routed $2m to support Colorado House candidate Manny Rutinel through Latino Victory Fund — and didn’t publicly announce it ahead of Rutinel’s victory

transformernews.ai

If you're at all interested in AI goings-on, we can't go on enough about how tremendously convenient and interesting our weekly newsletter is. One big story this week is: have you noticed these academics getting sucked into AI labs? What's going on there? www.transformernews.ai/p/gpt-56-get...

Take a break from doomscrolling Bluesky, and come doomscroll our snappy-as-hell newsletter, where we've got all the best of AI discourse rounded up for you. www.transformernews.ai/p/gpt-56-get...

“Hey, my boss wants to do something on AI. That’s all the direction I’ve gotten." There's a bit of a gap between what congress should know and what it does know about AI. But there're strange reasons it's come about, and it isn't *impossible* to fix www.transformernews.ai/p/how-to-fil...

Hey look, the Transformer weekly newsletter is already here! Just click through this link to see everything hope-inspiring, frightening, and bizarre about AI right now. The big story this week: might all this Mythos chaos be helpful in the long run? www.transformernews.ai/p/fable-anth...

The Transformer weekly newsletter is here, and if you've been wondering what the hell has been going on with all this Mythos stuff, we've been kind enough to break it all down into easily consumable bulletpoints. There's more through the link: www.transformernews.ai/p/fable-anth...