Are Astra and GPT-6.1 Sol less monitorable because they use recurrent depth?

Are Astra and GPT-6.1 Sol less monitorable because they use recurrent depth?

On Oct. 5, Microsoft’s Azure page for GPT-6.1 Sol said it ran “two inference passes instead of three.” Two hours after it was publicized, it was gone.

On Oct. 5, Microsoft’s Azure page for GPT-6.1 Sol said it ran “two inference passes instead of three.” Two hours after it was publicized, it was gone.

Philip Dowdell

·

Three states passed AI transparency laws — but who knows which models are actually covered?

California’s AI transparency law requires disclosures for the most powerful models. But it never requires companies to say which models hit the thresholds. New York and Illinois didn’t fix this gap either.

Samantha Stein

·

Three states passed AI transparency laws — but who knows which models are actually covered?

California’s AI transparency law requires disclosures for the most powerful models. But it never requires companies to say which models hit the thresholds. New York and Illinois didn’t fix this gap either.

Samantha Stein

·

Who is behind the AI safety backlash?

Following a viral resignation, the backlash against AI safety was sudden and swift. Various signs suggest an industry push is supporting it.

Tyler Johnston

·

Who is behind the AI safety backlash?

Following a viral resignation, the backlash against AI safety was sudden and swift. Various signs suggest an industry push is supporting it.

Tyler Johnston

·

OpenAI safety framework documents and GPT system cards illustrating California AI compliance concerns.

OpenAI wrote a safety framework for California. Its system cards are ignoring it.

Under SB 53, OpenAI’s Frontier Governance Framework calls for a loss-of-control risk tier for every covered model. The GPT-5.6 and GPT-6 Astra system cards assign none.

Brittney Gallagher

·

OpenAI safety framework documents and GPT system cards illustrating California AI compliance concerns.

OpenAI wrote a safety framework for California. Its system cards are ignoring it.

Under SB 53, OpenAI’s Frontier Governance Framework calls for a loss-of-control risk tier for every covered model. The GPT-5.6 and GPT-6 Astra system cards assign none.

Brittney Gallagher

·

OpenAI research papers beside a megaphone, illustrating concerns about research becoming propaganda.

OpenAI’s research is too important to be turned into propaganda

As we enter a new era of AI capabilities, OpenAI is facing increased pressure to spin its research in self-serving ways.

Tyler Johnston

·

OpenAI research papers beside a megaphone, illustrating concerns about research becoming propaganda.

OpenAI’s research is too important to be turned into propaganda

As we enter a new era of AI capabilities, OpenAI is facing increased pressure to spin its research in self-serving ways.

Tyler Johnston

·

Collage of matching pro-AI X posts illustrates suspected astroturfing linked to David Sacks.

A pro-AI, dark-money group tied to David Sacks appears linked to a new astroturfing campaign

Another round of suspiciously coordinated posting suggests a paid campaign to benefit Innovation Council Action, the $100 million advocacy group fighting for AI deregulation.

Tyler Johnston

·

Collage of matching pro-AI X posts illustrates suspected astroturfing linked to David Sacks.

A pro-AI, dark-money group tied to David Sacks appears linked to a new astroturfing campaign

Another round of suspiciously coordinated posting suggests a paid campaign to benefit Innovation Council Action, the $100 million advocacy group fighting for AI deregulation.

Tyler Johnston

·

Folder bearing the xAI logo, illustrating scrutiny of Grok 4.5 under California’s AI law.

SpaceXAI just released Grok 4.5. It may have broken California’s AI law in the process.

The potential violation by SpaceXAI, punishable by fines up to $1 million, will serve as an early test of whether California's new AI safety law actually has teeth.

Tyler Johnston

·

Folder bearing the xAI logo, illustrating scrutiny of Grok 4.5 under California’s AI law.

SpaceXAI just released Grok 4.5. It may have broken California’s AI law in the process.

The potential violation by SpaceXAI, punishable by fines up to $1 million, will serve as an early test of whether California's new AI safety law actually has teeth.

Tyler Johnston

·

Who We Are

Model Republic is a publication from The Midas Project, expanding our commitment to independent, public-interest research and advocacy. We examine how AI companies engage with policymakers, shape regulatory debates, and influence public decision-making.

Our Other Projects

Our Other Projects

01

The OpenAI Files

The OpenAI Files is the most comprehensive collection to date of documented concerns with governance practices, leadership integrity, and organizational culture at OpenAI.

Co-published with Tech Oversight Project.

Explore

02

Open Letter to OpenAI

More than 10,000 prominent AI experts, former OpenAI team members, public figures, civil society groups, and members of the public signed an open letter calling for greater transparency from OpenAI.

Co-organized with Encode and EyesOnOpenAI.

Explore

03

Watchtower

AI companies quietly edit their safety policies all the time — usually to weaken them. Watchtower tracks changes, announced or not, across major AI companies safety and security policies. If they won't keep a changelog, we will.

Explore

Smooth Scroll
This will hide itself!