AI Safety & Alignment

Today's Stories

1. 1

   Nvidia's Jensen Huang: develop AI "as fast as we can"

   Nvidia founder and CEO Jensen Huang told CBS News that AI development should move 'as fast as we can irrespective of anybody else,' despite recent calls for caution from Anthropic researcher Jacob Coxon and CEO Dario Amodei. The world's largest chipmaker is publicly rejecting a slowdown, setting up a visible split with those warning about AI risks.

   Huang also said Nvidia would never ship unsafe products, so the test is whether that pledge holds as development accelerates. Watch his full interview on Sept. 20 on 'Sunday Morning.'

2. 2

   iOS 27: Siri AI reads your apps by default; EFF shows how to limit it

   With iOS 27, Siri AI can read content from Apple apps by default, and the EFF outlines controls: disable "Show Content in Search" per app, opt out of "Improve Siri & Dictation," or switch back to "Siri Classic." By default, third-party apps are only searchable if their developers choose to add support, so what Siri AI can see depends on both developer choices and user settings.

   EFF says there is currently no way for you or app developers to block the on-screen awareness feature, so avoiding it may be the only protection for secure chats. Apple appears to be planning a single privacy toggle in a future update.

3. 3

   Citi CEO warns of 'Tsunami' of patching to secure AI defense at QEF 2026

   At QEF 2026, Citi's CEO said a 'tsunami' of patching lies ahead to secure AI defense, according to Bloomberg. This suggests banks and other enterprises may face a heavy, ongoing workload to keep AI systems defended, with security teams likely pulled into continuous patching.

   The comment hinges on how fast those patches are actually required and who inside large firms owns them; Bloomberg's report is the current source.

4. 4

   Panasonic Connect CIO 河野昭彦 to detail 2023-era gen-AI rollout at Oct 7 seminar

   Panasonic Connect group VP and CIO Akihiko Kono will give the keynote at ZDNET Japan's online security seminar on October 7 at 1 p.m., covering generative AI practices the group has pursued since 2023. The session frames shadow AI — generative AI adopted without security or governance approval — as a serious threat, and offers one large manufacturer's approach to rolling AI out safely.

   The talk is a company account, not an independent audit, so the test is whether Panasonic's culture-building examples translate to other organizations. The seminar runs online from 1 p.m. on October 7.

5. 5

   Dario Amodei's AI "pacing" plan split the industry in days, critic says

   Holistic AI co-CEO Emre Kazim published a critique of Dario Amodei's essay "We Must Pace the Frontier," noting the proposal split the industry into three camps within days of publication. The speed of that disagreement suggests coordination among labs and governments cannot be the foundation of AI safety, since the proposal already fractured the field it depends on.

   Kazim argues governance must instead operate at machine speed, with independent institutions verifying and AI systems monitoring; watch whether enterprises make governability an architectural requirement.

6. 6

   HPMOR reader says two years since university, a year and a half into rationalism

   A reader who finished HPMOR calls it the best novel they had read in more than a decade, and says they only began engaging seriously with rationalism about a year and a half ago. The reader says they only wished they had found HPMOR sooner, and describes their path into rationalism as beginning just a year and a half ago.

   The account hinges on self-reported details of when this reader found HPMOR and began engaging with rationalism, not on any verified timeline.