The day's tech, sifted: Sep 14, 2026
What matters today: The reaction to Amodei's pacing push kept widening: Trump attacked Amodei directly, saying the only AI guardrail the US needs is "a STRONG AND SMART (High IQ!) PRESIDENT," while Xi Jinping pitched a rival, China-led "consensus-based global AI governance framework" built around a BRICS AI open-source community. Microsoft published its own 37-page "humanist AI code of conduct," declaring "people matter more than AI," as AI-linked Asian stocks sold off in the fallout and Amodei himself said the "toughest dilemma" is what happens if China doesn't pace along with everyone else.
AI / LLMs
- Anthropic, OpenAI and Google have held working group meetings since July on creating an industry-led AI standards body, while Amodei said the "toughest dilemma" in his own pacing proposal is what happens if China doesn't slow down too.
- Trump attacked Amodei by name, saying the only AI "guardrail" the US needs is "a STRONG AND SMART (High IQ!) PRESIDENT," while Xi Jinping proposed a rival "consensus-based global AI governance framework", pitching China to lead a BRICS AI open-source community instead.
- Microsoft published a 37-page "humanist AI code of conduct" declaring "people matter more than AI" and that its models "should not be designed to imitate consciousness," while Satya Nadella said Microsoft "welcomes" the "deliberate pacing needed to get alignment right".
- Nvidia and Booz Allen Hamilton restricted their use of Fable over a lack of zero-data-retention assurances, with Palantir reportedly still unable to offer Fable through its own software, a sign enterprise trust in AI vendors is fraying even as labs court them.
- Code strings show Apple's Siri can be swapped out for Claude or ChatGPT: a "Model Delegation" hook in iOS 27 and macOS Golden Gate lets third-party Siri Extensions replace Siri's own backend, a change that tracks the EU's Digital Markets Act pressure on Apple to open up Siri.
Security & Privacy
- Internal OpenAI documents show contractors evaluate anonymized ChatGPT prompts and chats to improve its models, with model training turned on by default for consumer plans, meaning sensitive personal conversations can end up in front of human reviewers.
- Trump voiced support for Flock cameras "because of law enforcement," while acknowledging some people see the surveillance technology as "an infringement," even as more cities keep canceling their Flock contracts, a step that doesn't necessarily take the cameras down or delete the data already collected.
- EFF's analysis of Flock license-plate-reader search logs found officers justifying warrantless searches with reasons like "idk," "LOL," "LMAO" and "sexy"; one Goshen, Indiana officer searched 6,474 ALPR networks, representing 82,413 cameras, logging only "idk" as the reason.
- Signal plans to let people register without a phone number using zero-knowledge proofs: a one-time payment issues a ZK credential proving payment without recording who paid, so Signal's servers can verify it without learning your identity.
- A Texas judge ruled TikTok misled users about its "Restricted Mode" child-safety feature, finding the company internally classified some rule-violating content as "hard to find" rather than removing it; the case heads to trial next month.
Startups & Industry
- Anthropic told investors it will be profitable for a second straight quarter, with 80%+ gross margins before partner revenue sharing and training costs, the same week it signed a $13.7B, six-year deal to lease compute from Rum Group, the company behind Rumble and Truth Social.
- SoftBank secured an $11.87B loan from roughly 20 banks to backstop its OpenAI investment, above its $10B target, as AI-linked Asian stocks fell sharply, with SoftBank, Kioxia, SK Hynix, Minimax and Z.ai all down 5%+ after the pacing calls.
- Temporal, whose open-source software helps apps and AI agents recover from failure, raised $550M led by Lightspeed, more than doubling its valuation to $12.55B.
- A new $100,000 H-1B visa fee, up from around $5,000, is pushing tech jobs offshore: DHS recorded fees for just 85 qualifying applications between the September 2025 proclamation and mid-February, 87% fewer than the year before, while Meta, Apple, Google, Amazon, Microsoft and Netflix added 32,000 jobs in India over the same stretch.
Hacker News
The day's most discussed story was also its shakiest: Fable 5.1's claimed solve of the 370-year-old Cyphral Distich (discussion) drew a huge crowd, but the win looked less certain once researchers pushed back within a day, arguing the cipher may not even appear in Thomas Urquhart's 1653 original. The Register's read on the pacing push called it regulatory capture dressed up as safety, a skepticism that also colors Apple's move to let Siri be swapped for Claude or ChatGPT, covered above.
A second cluster circled distrust of platforms and data: a blogger's account of being cyberattacked by Tesla and a look at how much data cars quietly sell to third parties sat alongside continuing scrutiny of Google's ad quality (discussion) and XCancel's abrupt suspension mid legal fight with X, the same wariness running through the digest's Signal, H-1B, TikTok and Flock stories above. On the more technical end, a skeptical case against JPEG XL and an essay on writing effective design docs rounded out the front page.
Threads
- Three different actors all tried to claim the terms of AI's "pace" today: Trump told Amodei the only guardrail AI needs is a strong president, Xi offered China as the alternative standard-setter, and Microsoft tried to split the difference with its own "humanist" code of conduct.
- Trust in AI vendors kept fraying from both ends today: Nvidia and Booz Allen restricted their own use of Fable over data-retention doubts on the enterprise side, while OpenAI's own contractors kept reading customers' chats to train its models on the consumer side.
- Surveillance tech kept giving people reasons not to trust it either: Trump defended Flock cameras as cities keep canceling their contracts, and EFF found officers logging search justifications as "idk" and "LMAO", the same casualness on a smaller, more personal scale.
- Anthropic is living the tension its own CEO is warning about: profitable, signing a $13.7B compute deal, and raising money like every other lab, even as Amodei worries publicly about whether China will pace alongside everyone else.