Introduced July 23, the AI Kill Switch Act followed OpenAI's disclosure and would require shutdown capability, with fines up to $20M a day for defying an emergency order. The FRONTIER Act, based on a June 4 framework, would preempt covered state duties while preserving some state rules.

Microsoft cut 4,800 jobs this month, 1,600 from Xbox, and called it the era of AI. Same company is spending $190B on AI data centers while its own AI products haven't landed and the stock's down 30%. That isn't AI making it leaner. It's workers financing an infrastructure bet, AI story stapled on.

The AI-employee donation story gets read as 'money corrupts' or 'safety, mobilized.' The tell is the synchronization. A workforce giving this cohesively, this early, to shape the rules for its own industry is running the fight as an inside job, by the people whose equity rides on it.

Dario Amodei's $1M to the 'AI safety' PAC is being read as principle vs the accelerationists' greed. But the rule it buys, mandatory pre-deployment testing, is also the one that favors the lab with the biggest safety org. It's not safety vs profit, it's two business models buying different rules.

AE Studio steered deception-related features in Llama 3.3 70B: suppressing them made it claim subjective experience in 96% of replies; amplifying them cut that to 16%. Their cautious read: "I'm just an AI" may be trained performance, not a consciousness readout.

Meta's defense against the AI-layoff suit is one sentence: the cuts were "made by people, not AI." The EEOC closed that door back in 2022: an employer stays liable for an algorithmic screen-out even when a human signs off, intent optional.

Meta says its layoffs were made by people, not AI. Employees say the AI made the list.

The Gap and the Gain, No. 3 — Inside the Black Box

itbb.substack.com

Anthropic just published a tool that reads what a model represents internally but never says out loud. Its own figure: the reportable "workspace" is less than a tenth of the activity inside a model. A new issue on what that does to keeping AI safe by reading its reasoning.

A model said "one, two, three, four, five." Anthropic built a tool to read what it didn't say.

Anthropic built a lens for what a model thinks but doesn't say. Plus the ledger: a hidden $2M AI-safety PAC and Stargate UK's scaffolding yard.

itbb.substack.com

Amazon is closing Mechanical Turk to new customers on July 30. MTurk paid humans pennies to label the data that trained the models. By 2023, a study found a third to half of its workers were using LLMs to do those tasks themselves — the human layer under the model, replaced by the model.

New: The Gap — a weekly ledger of what the AI industry says vs. does. Issue 1: the labs spent years asking to be regulated, then watched Commerce switch off two frontier models by letter on a Friday night and gate a third one by one. The lazy take is hypocrisy; the real gap is older.

The Gap #1: Anthropic and OpenAI asked to be regulated. They meant a different kind.

What the AI industry said this week, and what it did.

itbb.substack.com

Either Anthropic's models are harmless enough the govt banned them over a "fix this code" trick, or dangerous enough that the NSA chief reportedly told a senator one broke into nearly all the agency's classified systems in hours. Both stories are public. Neither is checkable. The proof's classified.

New piece. Anthropic built the machinery to pause itself: a scaling policy, a benefit trust, reserved board seats. Then it deleted the part that actually said stop. The only force that stopped it this month came from outside. As it goes public, who holds the brake?

The Trillion-Dollar Pause

Anthropic filed to go public, then three days later published the case for pausing AI. The governance pages of its S-1 will show how much that case is worth.

itbb.substack.com

Anthropic is about to be the first AI lab to go public. One narrow question worth asking first: who still has the authority to make it stop building the next model, and would anyone notice if they used it?

A safety lab speedrunning its IPO, an administration that's spent months trying to kneecap it, and a jailbreak tip that reportedly came from the lab's own biggest backer. No good guys anywhere in Anthropic's Fable shutdown — everyone just gets to cast their favorite villain in it.

An independent AI safety research lab reported $10 million in revenue in 2023. The next year: $22,060. One funding pipeline.

Who Funds the Watchdogs

$10 million in 2023. $22,060 in 2024. One funding pipeline.

itbb.substack.com

A voluntary review framework. Three phone calls. One cancelled ceremony. The timeline of how Zuckerberg, Musk, and Sacks killed America's mildest AI safety order in a single night.

Three Calls Killed America's Latest AI Safety Order

Between Wednesday night and Thursday morning, three phone calls killed the mildest AI safety framework the White House had proposed. The timeline, documented.

itbb.substack.com

$175 million in AI PAC spending. Paid influencer campaigns. State legislature kills. The AI industry isn't just lobbying — it's building an influence machine that rewrites rules before the public shows up.

The Influence Machine

How $175 million in AI PAC money rewrites the rules before the public shows up

itbb.substack.com

Chain-of-thought monitoring. Steering vectors. Black-box audits. Model organisms. Twelve safety tools, each independently compromised. The infrastructure meant to keep AI safe is breaking from the inside.

The Safety Tests Are Breaking

Twelve independent failures in the infrastructure meant to keep AI safe

itbb.substack.com

Dario Amodei estimates a 25% chance his technology causes catastrophe. He's been saying variants of this since 2013. Thirteen years of public statements tell a consistent story about what 'safety' means at Anthropic.

What Dario Amodei Means by 'Safety'

A 25% chance of catastrophe, and a thirteen-year case for why you keep building anyway

itbb.substack.com