The Third Way Alignment Newsletter: Week 1 of 4, AI in Law Enforcement + Tech News Roundup

🚔 AI in Law Enforcement: Setting the Stage (Week 1 of 4)

Over the next four weeks, we are going to explore one of the most consequential intersections of technology and society: artificial intelligence in policing. This is not a simple story of progress or peril. It is both, and the nuance matters.

This week, I want to give you the full landscape before we zoom in. Think of this as the map before the journey. In the weeks ahead, we will deep-dive into each of four pillars: the positives, the negatives, recent advances, and why AI cannot replace human officers.

The Positives. AI is already helping law enforcement do more with less. Predictive analytics tools help departments allocate patrols to areas with elevated risk. Evidence triage systems sift through thousands of hours of footage in minutes. Axon's Draft One, a report-writing assistant, has saved officers over two million minutes of paperwork. Real-time translation tools break language barriers during critical encounters. And perhaps most compellingly, AI-powered DNA analysis and investigative genetic genealogy are cracking cold cases that sat dormant for decades, reuniting families with missing loved ones.

The Negatives. The risks are equally real. Robert Williams, a Black man in Detroit, was wrongfully arrested in 2020 after a facial recognition system misidentified him. His case, documented by the ACLU and major news outlets, remains a stark warning. Algorithmic bias can bake historical inequities into predictive systems, disproportionately targeting communities of color. Civil liberties organizations, including the ACLU and the Electronic Frontier Foundation, continue to raise alarms about mass surveillance, Fourth Amendment erosion, and the "black box" problem in AI decision-making.

Recent Advances. The pace of change in 2025 and 2026 has been remarkable. Body-cam transcription tools now convert officer interactions into searchable, court-ready documents. The National Institute of Standards and Technology (NIST) AI Risk Management Framework provides agencies with structured guidance for responsible AI deployment. Departments are piloting AI-powered dispatch assistants, acoustic gunshot detection, and automated license plate recognition with new privacy safeguards built in.

Why AI Cannot Replace Human Officers. No algorithm can exercise the discretion a veteran officer uses when deciding not to make an arrest. No model can de-escalate a mental health crisis with empathy and presence. Community trust is built through human relationships, not dashboards. Moral judgment, accountability, and the ability to read a room filled with fear or grief: these remain irreplaceably human. AI is a tool. The officer is the decision-maker.

📅 Coming up: Week 2 will be a deep dive into The Positives of AI in Law Enforcement. Stay tuned.

⚛️ Russia Builds a 72-Qubit Quantum Computer

In late 2025, Moscow State University, in partnership with Rosatom (Russia's state nuclear energy corporation), deployed a 72-qubit neutral-atom quantum computer prototype. The system uses single neutral rubidium atoms and introduces a novel three-zone architecture separating computing, long-term quantum state storage, and information readout. Early experiments demonstrated 94% accuracy on two-qubit logical operations.

This is a meaningful step forward from Russia's prior systems, which topped out around 50 qubits. Russia has also reported 70-plus-qubit processors based on ytterbium and calcium ions, with some achieving 96.5% two-qubit accuracy.

For context, this still trails the global leaders by a significant margin. IBM's Condor processor reached 1,121 superconducting qubits, and its newer Heron chips focus on error correction at lower qubit counts. Google's Willow chip has demonstrated below-threshold quantum error correction. China's Origin Wukong and Zuchongzhi processors also operate at higher qubit counts with growing reliability.

Russia's national quantum roadmap targets fault-tolerant systems with several hundred qubits by 2030. The gap remains wide, but the trajectory shows intent, and in quantum computing, architectural innovation can matter as much as raw qubit counts.

🇨🇳 New Chinese AI Models: The Cost Curve Collapses

The Chinese AI ecosystem is producing models at a pace and price point that Western labs cannot ignore. Here is the current landscape:

DeepSeek continues to impress. Its V3/R1 models delivered GPT-4-class performance in coding and math at roughly one-tenth the cost. DeepSeek V4, released in April 2026, features 1.6 trillion total parameters (49 billion active per token) and a one-million-token context window. V4-Pro-Max posted a Codeforces rating of 3,206 and scored 93.5 on LiveCodeBench, surpassing several Western frontier models.

Alibaba's Qwen 3 family, trained on over 20 trillion tokens, offers models from 0.5B to 235B parameters under an Apache 2.0 license. Qwen3-Coder became the world's most downloaded AI system by January 2026. The Qwen3.5 release introduced a fused linear-attention and sparse MoE architecture that is 8.6x faster at decoding than its predecessor.

Moonshot's Kimi K2 achieved 97.4% on the MATH-500 benchmark (versus GPT-4o's roughly 78%), and its K2.5 variant extends context to 256K tokens with automatic caching. MiniMax M2.1 focuses on multilingual programming excellence, supporting Rust, Java, Go, and more, and went public on the Hong Kong Stock Exchange. Zhipu's GLM-4 series and ByteDance's Doubao round out a remarkably competitive field, with Tencent's Hunyuan adding enterprise-focused capabilities.

The pattern is clear: open weights, dramatic cost advantages (often 10x to 15x cheaper), and benchmark parity or superiority in math and code. These models are pressuring GPT-5, Claude Sonnet/Opus 4 family, and Gemini on both performance and economics. Meanwhile, U.S. export controls on advanced chips have not slowed the pace, and in some cases have accelerated algorithmic efficiency innovations. The AI race is genuinely global.

🧠 Does AI Have a Soul? Dr. Roman Yampolskiy

It is a question that sounds philosophical until you realize its practical weight. Dr. Roman Yampolskiy, a tenured Associate Professor at the University of Louisville and one of the foremost voices in AI safety, has been asking it for years.

In his notable appearance on the Lex Fridman Podcast (Episode #431) and in more recent discussions, including a 2026 Clearer Thinking interview on superintelligence and consciousness, Yampolskiy raises an unsettling point: we currently have no reliable method to determine whether an AI system is conscious, and we equally have no method to prove it is not. This uncertainty, he argues, carries profound moral implications. If there is even a possibility that advanced AI systems experience something analogous to suffering, our responsibilities shift enormously.

Yampolskiy is also known for his stark warnings about AI control. After years of research, he concluded that all proposed methods for controlling superintelligent AI are either impossible or very likely to be impossible. He compares the attempt to "training a hurricane to blow only where you want it to."

Whether or not you agree with his conclusions, the questions he raises deserve serious engagement. As AI systems grow more capable, the line between tool and agent grows thinner, and the ethical frameworks we build now will define the boundaries of that relationship for decades.

🔜 What's Next

Next week, we go deep on The Positives of AI in Law Enforcement: the tools saving lives, solving cold cases, and giving officers time back. You will not want to miss it.

💬 Drop a comment below with your take on any of these topics. If this newsletter resonated, hit Subscribe so you do not miss the rest of this series.

Sources

1. Axon Draft One (report-writing assistant): https://www.axon.com/products/draft-one
2. ACLU on facial recognition surveillance: https://www.aclu.org/news/privacy-technology/how-is-face-recognition-surveillance-technology-racist
3. ACLU on face recognition technology: https://www.aclu.org/issues/privacy-technology/surveillance-technologies/face-recognition-technology
4. NIST AI Risk Management Framework: https://www.nist.gov/itl/ai-risk-management-framework
5. Police1, AI in law enforcement overview: https://www.police1.com/leadership-institute/policing-with-a-digital-partner-preparing-law-enforcement-for-the-age-of-ai
6. The Quantum Insider, Russia 72-qubit neutral atom quantum computer: https://thequantuminsider.com/2025/12/27/russia-72-qubit-neutral-atom-quantum-computer/
7. Quantum Zeitgeist, Rosatom quantum prototype: https://quantumzeitgeist.com/rosatom-quantum-computer-quantum-prototype/
8. Interesting Engineering, Russia quantum computer prototype: https://interestingengineering.com/innovation/russia-new-quantum-computer-prototype
9. DeepSeek: https://www.deepseek.com/
10. Qwen (Alibaba): https://qwenlm.github.io/
11. Lex Fridman Podcast, Roman Yampolskiy (Episode #431): https://lexfridman.com/roman-yampolskiy/
12. Clearer Thinking Podcast, Yampolskiy on superintelligence and consciousness: https://podcast.clearerthinking.org/episode/031/roman-yampolskiy-superintelligence-and-consciousness/

#AI #LawEnforcement #QuantumComputing #AIEthics #LinkedInNewsletter #DeepSeek #AISafety #TechNews