All stories

Yesterday

Open models have real benefits. Do they outweigh the risks?

Key points
  • Anthropic's report argues Z.ai's GLM-5.3, released in August, poses a major cyberattack risk because anyone can download and modify it.
  • The top eight models on the Artificial Analysis LLM leaderboard are all closed, though open models reportedly trail by only a few months.
  • A Mozilla report found open models typically cost less, and running them locally can keep sensitive business data on in-house hardware.
  • Closed models get account shutdowns and misuse filters. Downloaded open models have none, and researchers reportedly found safety training easy to undo.

Why it matters for marketers: For marketing teams weighing AI tools, open models offer local control and lower costs, but fewer built-in misuse safeguards than closed services.

MIT Technology ReviewPolicy & safety

AI labs are buying data from dead startups to train their models

Key points
  • SimpleClosure added data sales in April; its Asset Hub has handled over 150 deals, expecting over $10 million returned to startups by end of 2026.
  • Google reportedly outbid Mercor for Spirit Airlines' data in August for $10 million, including 100 million emails and 500 million Teams messages.
  • Founders have little visibility into buyers and their plans, and proceeds typically go to senior creditors and investors rather than employees.
  • No federal law bans such sales, and an attorney calls the data 'legally radioactive' because of private details in emails and calendars.

Why it matters for marketers: Content and data teams should expect wind-down archives to be sold as AI training material, raising consent and licensing questions for any company's workspace data.

MIT Technology ReviewBusiness & funding

This week

Earlier