It’s Wednesday, September 9, and today’s issue is about the messy middle of the AI race: who gets credit when a model helps with discovery, who is accountable when AI touches stolen data, and why agents still feel more compelling in a demo than in a normal workday.

Plus: a new Real vs. AI challenge and our favorite tool of the day.

An NYU mathematician says OpenAI handled a career-making math problem unfairly, adding a human question to a machine-assisted result. The fight is about more than one proof: as models enter research, fields will need clearer rules for credit, review and responsibility.

A new report alleges Chinese hackers ran AI systems on information taken from a U.S. target. The case puts a familiar security problem in a new frame: stolen data becomes more useful when tools can search, sort and analyze it at scale.

Google’s Chrome browser will ship updates every two weeks as AI changes the pace of security work. Faster releases can close gaps more quickly, but they also demand that companies keep up with a browser that is changing more often.

ElevenLabs, the AI voice company, has hired a chief financial officer while it considers a 2028 public listing. The move shows voice AI is becoming a business with the same financing and governance demands as the larger model companies.

Agentic AI promises unlimited productivity. Delegate your most tedious tasks to bots and let them work while you sleep. So why aren’t more people using agents? It’s not because people aren’t using AI. Pew Research found that 60% of U.S. adults were already using LLMs in 2026—one of the fastest rates of technology adoption in history.

Agents are different. ChatGPT asks almost nothing of you beyond opening a browser and typing. Agents ask users to delegate: connect accounts, grant permissions, define workflows, supervise results, and trust software to act on their behalf. That turns out to be a much higher bar. → Read the full article here.

Mathematician Jacob Tsimerman is raising a difficult question as AI systems take on more advanced research: what does it mean to trust a result when the work is produced or accelerated by a model? In mathematics, a proof is not just an answer. It is a chain of reasoning that other people can inspect, challenge and eventually stand behind.

That makes the issue relevant far beyond math. If frontier systems begin contributing to science, engineering or policy, institutions will need ways to verify the work and assign credit without treating a model’s output as self-proving. Watch for universities, journals and labs to develop their own standards before AI-assisted research becomes routine. → Read the full article here. (Paywall)

A California investigation tied to a WeChat-linked hack points to a new cyber risk: once attackers get access to a large trove of data, AI can help them search it, organize it and identify the most valuable pieces faster. The hard part is no longer only keeping intruders out; it is limiting what a breach can become after the first break-in.

For companies and public agencies, that raises the value of basic controls such as data minimization, access limits and fast incident response. The next test will be whether security teams can spot AI-assisted misuse early enough to stop the data from being turned into a much more precise attack. → Read the full article here. (Paywall)

Mistral Raises $3.5B in a Record European AI Round: The French AI company secured a new $3.5 billion round.

Anthropic Reportedly Walks Away From Decart Deal: Anthropic is said to have stepped away from a potential $6 billion acquisition of Decart. (Paywall)

Slovenia Clears Tesla’s FSD Ahead of an EU Vote: Slovenia approved Tesla’s Full Self-Driving driver-assistance system.

Rep. Lori Trahan Pushes for AI Rules: Rep. Lori Trahan is pressing for a stronger federal response to AI.

Qualcomm and Amazon Strike a Data-Center Deal: Qualcomm reached an agreement with Amazon on data-center infrastructure.

Trae.ai is ByteDance’s AI-powered coding environment built on VS Code. It combines inline code completion, file-aware chat and Builder mode for planning and applying multi-file changes from a natural-language request.

Try it when you want to prototype a small project or make coordinated edits across a few files. → Explore Trae.ai in the Forward Future Tool Library

Thanks for reading. See you next time!

— Matthew Berman, Nick Wentz & the Forward Future Team