What happened
Sam Altman and Elon Musk backed Dario Amodei's call for a slowdown in model releases, while Marc Benioff at Dreamforce doubled Salesforce's bet on AI agents.
Why it matters
Even as top AI leaders urge caution, major software vendors are pushing agents into core products — a split between safety talk and commercial action that may widen.
What to watch
Whether Washington acts on AI rules, where Sen. John Kennedy said Congress cannot currently pass legislation; Anthropic's new metrics for monitoring AI development speed are the next concrete checkpoint.
WHO IT HITSEnterprise software buyers and IT teams evaluating agentic platforms face a confusing landscape: vendors such as Salesforce and Splunk are racing to ship agents, while security firms like Exaforce and Cohesity are simultaneously selling kill switches to contain them.
Summaries like this, in your inbox every morning.
The week's AI news split into two opposing currents. On one side, OpenAI's Sam Altman and SpaceXAI's Elon Musk joined Anthropic's Dario Amodei in backing a slowdown in model releases, following high-profile departures of researchers worried about runaway AI. Anthropic also released metrics for monitoring how fast AI is advancing. On the other side, Salesforce's Dreamforce conference saw the CRM company double its bet on agents as the next era of software, even as CEO Marc Benioff compared AI companies' approach to social media that 'hurt kids.' Splunk likewise rebuilt its platform around agents, and the AI Infra Summit drew hundreds of vendors pitching infrastructure.
The tension extends to Washington, where prospects for AI legislation look dim — Sen. John Kennedy's blunt assessment suggests no near-term action. Meanwhile, cybersecurity vendors launched a wave of agent kill switches and guardrail services, reflecting growing concerns about agent behavior in production.
The stakes hinge on whether the safety calls translate into actual slowdowns or remain rhetoric while enterprise adoption accelerates. For companies deploying agents, the practical question is whether guardrail tools mature fast enough to match the pace of agentification across CRM, security, and infrastructure platforms.
For example, today's edition would include:
AI-summarized, only the topics you pick — one digest a day via Email, Slack, or Discord.
Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →
Ask AI anything about this article. Q&As are published on this page for other readers too.
Eva Brucherseifer and Jan Muehlig will present 'What would it take?
Meta Platforms shares are up 24.34% over the past month, as Muse became the #1 app in the App Store one week a…
Broadcom, Meta Platforms, and Microsoft are all trading around breakeven for the year
A review of newspaper archives from 1919 to 1945 found striking parallels between early atomic-energy debates…
In the first posts from the new DeepMind Institute, researchers Rohin Shah and Anca Dragan argue visible chain…
Anthropic published an index scoring its own development work, and says 26 percent of it now sits at AL4 — Epo…
