A new preprint study of authoritarianism-enabling AI presents another set of evidence of the gaps in safeguards, writes Tim Bernard.

Technologies cannot typically be accurately described as intrinsically authoritarian. However, as many Tech Policy Press readers know, technology can, and is, used to advance authoritarian goals. The various forms of AI are no exception. In the US, there has been recent scrutiny of tools being developed for ICE to identify deportation targets, and longstanding issues have been identified with predictive policing systems that purport to identify where crime is likely to occur or who is likely to commit a crime. Emerging from a fully-authoritarian context, China’s DeepSeek LLM has been shown to adhere to its homeland’s communications policy regarding Taiwan and the 1989 events at Tiananmen Square.

As illustrated by the policing examples above, authoritarianism-enabling AI goes beyond LLMs and frontier models, and includes systems that have been in place for some time. What then makes AI of particular interest in this context? In a new preprint study, “From Democracies to Autocracies: How AI Systems Enable Authoritarianism by Design,” policy researchers Jeba Sania, Marta Ziosi, and Fazl Barez draw together common threads amongst AI systems of concern, noting that they include “specific features such as statistical learning, extensive data collection and analysis, and black-box algorithmic decision-making that resists explainability and thus evades effective oversight.”

Benedict Cumberbatch, Alan Cumming and Benedict Wong have urged the U.K. government to block the Paramount-Warner Bros. merger, saying it “threatens to inflict immense harm on the British public.”

In an op-ed addressed to culture minister Lisa Nandy, the trio said her intervention could be “prove one of the most important decisions any culture secretary has taken for UK film, television and media in a generation.”

Meta's $50 billion AI data center in Louisiana is facing scrutiny over secrecy, energy demands, public incentives and its impact on locals.

Meta’s massive artificial intelligence data center in rural northeastern Louisiana has become a symbol of the extraordinary infrastructure race unfolding behind the rapid expansion of AI, with the project drawing scrutiny over its scale, energy needs, financing arrangements and the secrecy surrounding the deal that brought it to the region.

The project, known as Hyperion, is expected to represent an investment of more than $50 billion and span nearly 10 million square feet, making it the largest data center in Meta’s global fleet. The facility is designed to provide 5 gigawatts of computing capacity for AI workloads, including the training and development of large AI models.

The development has transformed Richland Parish, a predominantly rural area in northeast Louisiana, as Meta and its partners build the infrastructure required to support increasingly power-intensive AI systems.

The scale of the project has also raised questions about how the deal was negotiated.

In the U.S. Market, everyone beat something in the June quarter, but investors only cared about what sat underneath the beat.

The market was not rejecting telecom, but it was separating valuable growth from expensive growth.

"That distinction reveals something much larger. The fastest-growing businesses in US telecom are increasingly built on infrastructure created for another purpose. Cable is selling mobile service through wholesale agreements and WiFi. Wireless carriers are selling home broadband through spectrum bought for smartphones. Verizon is selling dark fiber to hyperscalers building AI infrastructure."

The networks are still being built. AT&T alone plans to invest between $23 billion and $24 billion annually through 2028. But a growing share of the next dollar of revenue is coming from finding a second customer for an asset that already exists.

Telecom has discovered capacity arbitrage.

Paramount Skydance announced that it was halting its merger with Warner Bros. Discovery until July of next year after a lawsuit from Democratic states. Paramount is also facing a lawsuit from the Writers Guild of America, which argues that the merger would lead to job losses for screenwriters and journalists.

Paramount Skydance is run by David Ellison, whose father, Larry Ellison, is the billionaire founder of Oracle and a prominent ally and financial backer of President Trump. 

Rep. Tim Burchett (R-TN) introduced a bill that would ban the federal purchasing and federal funding of Flock cameras, which he calls “unconstitutional."

Part 1 video description

In December 2025, the Trump administration created a federal task force to challenge state AI laws in court. A Commerce Department list of "onerous" laws is due around March 2026. This video, Part 1, explains a law the fight targets: California’s Transparency in Frontier Artificial Intelligence Act, which does not ban any AI system or impose liability for outcomes, but requires the most powerful AI developers in the world to document how they assess catastrophic risk.

What Part 1 covers:

• Why SB 53 exists: how Newsom's veto of SB 1047 led to a working group from Stanford, the Carnegie Endowment, and UC Berkeley, and a trust-but-verify transparency framework. The bill cleared the state Senate 37-0 on first passage. The enacted version passed the Assembly 59-7 and the Senate 29-8 in September 2025.

• The compute threshold: what 10²⁶ floating-point operations means in human terms, why compute is the one variable the industry can't easily obscure, and how the two-tier test isolates roughly 5–8 companies.

• The frontier AI framework: the four questions large developers must answer before deploying a model, what catastrophic risk means under the law, and what is explicitly excluded.

• Model security: why a frontier model's weights are a specific concern, and what Meta's 2023 LLaMA leak showed about the stakes.

Part 2 covers the self-reporting problem, whistleblower protections, CalCompute, and the federal collision.

#sb53 , #CaliforniaAIlaw, #frontierai, #airegulation, #aisafety, #artificialintelligencepolicy, #calcompute, #trumpai, #trumpaiexecutiveorder, #californiatechpolicy, #aicatastrophicrisk, #beadbroadband

July 22, 2026 - Graham Media Group has opened a new coffee shop in Detroit that doubles as a community meeting center and a television studio connected to its local NBC affiliate.

The new space, called Fourgrounds, opened to the public on Wednesday. It operates as a traditional coffee shop while allowing the public to connect with reporters and other newsroom personnel at WDIV (Channel 4).

In a statement this week, Graham Media said the coffee shop will allow its local journalists to meet with members of the community to solicit story ideas and conduct news gathering outside the traditional newsroom setting.

Cheating as a business model exists because Americans can no longer sue big business. A Democratic Congress could fix this problem with one simple law. Plus, a defeat for Paramount-Warner, and more...

As usual, there’s a lot of monopoly-related news. There was an important win in the Paramount-Warner merger fight, the world of tech giants is in a civil war over open source AI models, and Lina Khan continues to annoy the superrich by becoming the Chair of one of the most important economic development posts in New York City.

Before getting to all of that in the full round-up I want to discuss a political opportunity to improve the lives of Americans very quickly.

In a few months, it’s likely that Congress will change hands. If it does, one thing Democrats could do is pass a law making it possible to sue big companies who cheat their customers, suppliers, or employees. Right now, it’s virtually impossible to do that, and the result is a degradation of our economic order so significant that we’ve had to invent a new word to describe it.

Our high‑altitude platform and phased‑array antenna deliver ubiquitous, high‑bandwidth, low‑latency coverage from the stratosphere, at a fraction of the cost of terrestrial networks and LEO satellite solutions.

Accessible anywhere, fueled by clean hydrogen, powered by World Mobile Chain.

Outdated rules tying carriers to copper phone lines are diverting billions from fiber and wireless buildout, and the FCC should finish modernizing them.

But there’s an obstacle to next-generation networks that too few realize: Legacy rules that bind carriers to the copper networks of yore. These rules require carriers to constantly reinvest in copper wires, dedicate largely empty buildings to 1980s technology, and prevent telephone traffic from being routed and exchange efficiently.

These rules are enormously expensive. AT&T alone spends more than $6 billion each year maintaining copper lines. And AT&T is only one example: 2,425 voice service providers incur unnecessary costs to interconnect with and support the legacy network. 

As data centres have grown in size and number in Australia, they have also become the subject of fierce public debate. So what’s the best way forward?

At first glance a data centre looks like a bland, boring warehouse. But these buildings, stacked with thousands of servers, are the beating heart of the internet and the booming artificial intelligence (AI) industry.

As data centres have grown in size and number in Australia, they have also become the subject of fierce public debate. State and federal governments, as well as the tech industry, are pushing for new data centres to be rapidly built in addition to the roughly 160 that already exist around the country.

How does geography impact AI performance? Our latest study examines how 5G latency varies across U.S. cities, exploring the role of infrastructure in cloud responsiveness.

Latency – the time it takes for data to travel from a user to a server and back – is becoming an important factor in AI performance. That’s because advanced AI services often do their calculations inside data centers hosting servers used by the likes of Google, Amazon Web Services (AWS), Oracle and Microsoft. Network latency is a part of the path necessary to extricate AI computations from a hyperscaler’s cloud and deliver them to an end user.

Further, time-sensitive AI applications – like real-time video analysis or voice chatting – will need low latency connections in order to quickly shuttle AI computations to the users (or robots or agents or whatever) who need them. As AI matures, these snappy connections may grow in importance.

But, as Ookla’s mobile 5G data shows, cloud latency can be impacted by a variety of factors including geographic location, internet routing infrastructure, and operators’ traffic steering, peering, and interconnect policies.

How a thirteen-year apprenticeship under McCarthy's chief counsel became the blueprint for a word doing very different work seventy years later.

One of the things I love most about research is that it rarely takes you exactly where you expect to go. I went looking for a rhetorical pattern and found a biography instead.

I had been tracking something simple at first: how often, and how casually, Donald Trump has started reaching for the word “communist” again.

Not “socialist.” Not “radical left.” Communist, specifically, applied to open borders, bail reform, transgender healthcare, opposition to voter identification, health-insurance policy, a sitting Minnesota governor and, in the same breath, to the very Democratic Party he also claims is secretly in the pocket of private insurers.

I wanted to know whether this was new. I wanted to determine whether it was strategic.

I did not expect to end up reading about a disbarred lawyer who died in 1986.

But that is where the thread led.

I wrote about the Whitehouse Ratepayer Protestor Pledge earlier today. Red Lake Nation News reports…

President Donald Trump announced a non-binding pledge by U.S. power producers and ‌data centers on Thursday to fund or build energy infrastructure that would meet massive AI-related power needs while shielding consumers from high electricity costs.

Consumer advocacy groups and administration critics, however, dismissed the pledge as an empty promise to consumers.

Below is Section 1 and Section 8 presented that way, plus a short list of draft-only concepts that were dropped. (Underlined words in the draft executive order; boldfaced in the final order.)

Ensuring a National Policy Framework for Artificial IntelligenceBy the authority vested in me as President by the

In an interview with the Economist’s Editor in Chief Zanny Minton Beddoes, Elon Musk said that China may become the world leader in AI. The only thing now preventing it is China’s lower computing power. But that could change.

China already has the big advantage in electricity capacity versus the U.S., a 3 to 1 advantage:

China could develop its chips and increase its compute, as the Wall Street Journal just reported on China’s ambitions. If it does, that’s when China may surpass the U.S., in Musk’s view.

On top of that, China has 50% of the world’s top AI researchers:

Explosive growth in energy-intensive AI data centers is outstripping the pace of power grid interconnection and transmission expansion. While operational flexibility has been proposed to mitigate this stress, existing processes are often reactive and evaluate projects only after they enter a multi-year interconnection queue.

To address this, we introduce a planner-initiated siting framework that integrates (i) reliability-gated screening, (ii) system-wide market-impact assessment under standardized flexibility envelopes (firm, pause, and shift), and (iii) entropy-weighted multi-criteria scoring to produce ranked, pre-certified catalogues of interconnection-ready locations. Applied to a synthetic 2000-bus Texas power system, the framework demonstrates that operational flexibility expands the siting frontier by 9-17% at 1 GW and 19-21% at 2 GW compared to firm operation.

Median all-hour average prices remain essentially unchanged ($24.32/MWh for the 2 GW cases), and the shift envelope attenuates peak-hour price dispersion by approximately 3.4% with minimal side effects during off-peak hours. Utilizing pre-certified envelopes to bypass major transmission reinforcements, this workflow enables first energization in 12-18 months-a conservative reduction of 3.5-4 years versus the conventional 5-8 year project-led process. This technology-agnostic framework provides a proactive decision-making tool for system operators and regulators to fast-track large flexible loads while preserving grid reliability and market stability.

The study presents a planner-initiated framework for siting data centers, integrating reliability screening and market-impact assessment. It expands feasible locations by 9-21% when accounting for operational flexibility and reduces interconnection time by 3.5-4 years, while maintaining stable prices and reducing peak-hour price dispersion.