AI Watch
The race to ship artificial intelligence moves faster than the scrutiny that should follow it. AI Watch slows things down. Each report here looks at one model, company, or policy and asks the questions the launch posts skip: what does it actually do, who is affected when it goes wrong, and what does the documented record show?
We have no sponsorship relationship with any AI company. Where a tool earns praise, we note it; where the concerns are real, we lay out the specifics with sources. Select a report below to read the full analysis.
deployment guides
industry impact
AI Coding Assistants 2026: Choosing the Right Tool
AI coding assistants have transformed software development with GitHub Copilot, Cursor, Codeium, Tabnine, and Continue offering different approaches to AI-assisted programming. Productivity studies show 30-50 percent acceleration on common tasks. The tools differ in privacy practices, model selection, IDE integration, and pricing. Choice depends on workflow preferences, privacy requirements, and budget.
Read analysis →AI Energy Crisis: Data Centers and Grid Strain
AI training and inference are driving unprecedented data center power consumption. Goldman Sachs projects data center electricity demand to grow 160 percent by 2030, primarily driven by AI workloads. Major tech companies are signing nuclear power contracts to secure capacity. Some grid operators warn AI growth threatens reliability. The climate implications conflict with corporate sustainability commitments.
Read analysis →AI in Legal Industry: Paralegal Roles Transforming
AI tools are transforming legal work traditionally performed by paralegals including document review, legal research, contract analysis, and case summarization. Major law firms report 60-80 percent reduction in document review time using AI tools. Paralegal hiring has slowed significantly as AI handles routine tasks. The profession is shifting toward AI oversight and complex work that requires human judgment.
Read analysis →AI vs Doctors: Where Algorithms Already Outperform
AI diagnostic tools have demonstrated accuracy matching or exceeding human physicians in several specific tasks including dermatological cancer screening, diabetic retinopathy detection, ECG interpretation, and certain cancer diagnoses. The 500+ FDA-cleared AI medical devices represent rapid technology integration into clinical practice. The technology augments rather than replaces physicians, with best results from human-AI collaboration.
Read analysis →AI vs Radiologists: When Algorithms Read Scans Better Than Doctors
AI systems have demonstrated accuracy matching or exceeding board-certified radiologists across multiple imaging modalities including mammography, chest X-rays, and CT scans. Google Health's AI achieved 11.5 percent fewer false positives and 9.4 percent fewer false negatives than human radiologists in breast cancer screening. The technology is not replacing radiologists entirely but is fundamentally reshaping the profession from primary image interpreters to AI-assisted diagnosticians.
Read analysis →model releases
Claude Opus 5: Anthropic's Latest Frontier Model
Anthropic released Claude Opus 5 in early 2026 as the latest in its frontier model series, demonstrating significant improvements in coding, reasoning, and long-context analysis. The model maintains Anthropic constitutional AI safety approach and was developed under the company Responsible Scaling Policy. Opus 5 sets new state-of-the-art results on multiple coding benchmarks and offers extended thinking capabilities for complex problems.
Read analysis →Gemma 4 Deep Dive: Open AI That Rivals Closed Systems
Google released Gemma 4 as a 26 billion parameter Mixture of Experts model with a 256,000 token context window under an Apache 2.0 license. The model challenges the assumption that competitive AI requires closed proprietary systems. Gemma 4 runs efficiently on consumer hardware with only 8 billion active parameters per inference, making it deployable on devices with 16GB of RAM. The release signals a strategic shift in Google's approach to open AI development.
Read analysis →GPT-5: The Next Frontier Model From OpenAI
OpenAI continues development of GPT-5 with reported training compute approaching 10^26 FLOPS, significantly exceeding GPT-4. The model is expected to demonstrate improved reasoning, longer context, multimodal capabilities, and potentially agent-like behavior. Release timeline has slipped from initial 2024 expectations to 2026 amid safety evaluation requirements and infrastructure constraints. The model will face EU AI Act systemic risk obligations from launch.
Read analysis →privacy implications
safety governance
AI Alignment Research 2026: Approaches and Open Problems
AI alignment research focuses on ensuring AI systems behave in accordance with human values and intentions. Major approaches include reinforcement learning from human feedback (RLHF), constitutional AI, scalable oversight, interpretability research, and safety evaluation. Progress remains challenging as model capabilities outpace alignment techniques. Core open problems include reward hacking, deceptive alignment, and aligning superhuman systems.
Read analysis →AI Hiring Tools and Algorithmic Discrimination
AI hiring tools used by employers including resume screening, video interview analysis, and skills assessment platforms have repeatedly demonstrated discriminatory outcomes. Amazon famously scrapped an internal AI recruiting tool that downgraded women candidates. HireVue facial expression analysis has been challenged for bias. The EEOC has issued guidance that AI hiring tools must comply with anti-discrimination law, but enforcement remains limited.
Read analysis →Deepfake Detection 2026: Tools, Accuracy, and Limitations
AI deepfake generation has outpaced detection capabilities, creating a fundamental verification crisis for digital media. Current detection tools achieve 75-90 percent accuracy on known generators but fail on novel methods. Major platforms have deployed automated detection but face an arms race where each detector improvement is followed by generator improvements. The C2PA content provenance standard offers a different approach through cryptographic signing rather than detection.
Read analysis →EU AI Act: Complete Enforcement Timeline and Deadlines
The EU AI Act implements the world's first comprehensive AI regulation through a phased enforcement timeline stretching from 2025 through 2027. Prohibited practices including social scoring and most real-time biometric identification took effect in February 2025. General purpose AI model obligations started in August 2025. High-risk system requirements become fully applicable in August 2026. Maximum penalties reach 35 million euros or 7 percent of global annual turnover.
Read analysis →Make your voice heard
Free to get started. No credit card required.
Join Open Public Voice