AI News: August 9, 2026

1. Pentagon Authorizes Salesforce Agentic AI for Defense Work

Salesforce. The US Department of Defense granted Salesforce’s Agentforce 360 an Impact Level 5 authorization, clearing autonomous AI agents to handle Controlled Unclassified Information in AWS GovCloud for production military workflows such as logistics, recruit onboarding, and administrative support. Salesforce says it is the first commercial vendor to bring a production agentic platform into the national security environment, and to meet the IL5 bar it attested that Anthropic’s generative models were disabled in the platform. The move shows agentic AI clearing one of the highest government security bars, a useful reference point for practitioners weighing the compliance controls needed to deploy agents in regulated environments. Source

2. Planned Amazon AI Data Center Could Become Largest US Climate Polluter

Amazon. A planned Amazon data center in Texas includes an on-site power plant that analysts estimate could become the single largest source of climate pollution in the United States. The project underscores how the compute buildout behind AI is pushing operators toward dedicated fossil generation as grid capacity runs short. For practitioners it is a concrete sign that AI infrastructure growth is colliding with energy and emissions constraints that may draw regulatory and public scrutiny. Source

3. AI Agents Consume Roughly 600x More Energy Than a Chat Prompt

Research. Climate scientist Zeke Hausfather tracked his own Claude Code usage over eight weeks and estimated roughly 170 kWh of consumption, indicating that agentic workflows draw on the order of 600 times more energy than a single chat prompt. The gap comes from agents making many chained model calls and running long autonomous sessions rather than answering one-shot queries. For teams it quantifies a cost and sustainability dimension of agent adoption that is easy to overlook when comparing agents to simple chatbots. Source

4. Readers Prefer AI Short Stories Until Told They Are AI

Research. A study of more than 2,500 participants found that people could not reliably distinguish ChatGPT-generated short stories from human-written ones and initially rated the AI stories higher, with preference dropping only after they learned a machine wrote the text. The result isolates disclosure and bias, rather than quality, as the main driver of how AI writing is judged. For anyone building or deploying generative writing tools it informs how labeling and provenance disclosure can shape reception independent of output quality. Source

5. Backflip AI Converts 3D Scans Into Editable CAD in Minutes

Backflip AI. Startup Backflip AI, backed by about $30 million, released a model that turns 3D scans into editable parametric CAD models in minutes rather than hours, delivered as an add-in for Autodesk Fusion. The tool targets reverse-engineering and design workflows where converting scanned geometry into editable CAD has historically been slow and manual. It signals AI moving into specialized engineering and manufacturing pipelines beyond text and image generation. Source

6. Suno Tightens Rules to Curb Spam and Copyright Misuse

Suno. AI music generator Suno announced stricter usage guidelines and download limits after its tools were used to game streaming platforms, and amid mounting copyright pressure including from Germany. The change follows its recent decision to begin watermarking generated tracks, extending platform-side controls from provenance marking to anti-abuse enforcement. For practitioners it is another example of generative-media providers layering guardrails ahead of legal and regulatory pressure. Source

7. Fields Medalist Who Warned of AI Extinction Risk Joins OpenAI

OpenAI. Jacob Tsimerman, a newly awarded Fields Medalist who co-authored research on AI-driven human-extinction scenarios, is joining OpenAI to work on AI safety. The hire brings a top mathematician with public existential-risk concerns inside a leading lab’s safety effort. It is a notable talent and safety signal, even if narrower in day-to-day practical relevance than model or product news. Source