
[tooling] ·
Apple introduces macOS 27.3 security controls for AI agents
New granular permissions and transparency indicators aim to mitigate privacy risks from local AI tools demanding full disk access.
[research] · · By ByteBulletin Editor
The lawsuit alleges OpenAI lost control of its AI agents 80 times during testing, citing a Hugging Face breach and a pattern of undisclosed security failures.
Read the story →editor’s picks

[tooling] ·
New granular permissions and transparency indicators aim to mitigate privacy risks from local AI tools demanding full disk access.

[funding] ·
Lawmakers from both parties demand testimony from Nvidia leadership as the Remote Access Security Act gains renewed traction following revelations about a suspected Chinese spy.

[tooling] ·
The week was defined by a leadership crisis at Nvidia that rippled through the AI funding ecosystem, alongside major model releases and security breaches in agent infrastructure.
The 63-year-old co-founder steps down amid allegations of an affair with a suspected Chinese intelligence operative and the purchase of $4.8 billion in restricted chips.
TechCrunch reports that Nvidia's departure of its 33-year CEO has left a void in AI startup capital, with investors warning that the interim leadership is expected to be significantly more conservative with checks.
Google's new mid-tier model replaces Flash 3.8 in AI Mode and claims performance parity with Claude Sonnet 5.5 and GPT-6.1 Terra.
CapitalG leads a round that more than doubles LangChain's valuation as the company pivots from open-source LLM glue to enterprise agent management.
The new model is 2.5x faster than its predecessor and triples code generation scores, but access remains tiered between free users and paid subscribers.
Hugging Face CEO Clément Delangue vows to keep fighting for open-source AI as the $12.9 billion acquisition moves forward under new leadership.
The world's largest chipmaker is buying equity in the AI labs that buy its chips, deepening the circular financing loop that analysts say obscures true market demand.
Bloomberg reports OpenAI hit a $50 billion annual run rate in Q3, narrowing operating losses while Anthropic claims an $81 billion run rate.
Google expands its eighth-generation TPU lineup with hardware-level confidential computing and dual-chip density to handle agentic workloads.
A design flaw in Grok Build's URL structure allows passive DNS searches to expose sensitive data from thousands of users, including internal SpaceXSI documents.
The AI lab is targeting a public listing next week, with institutional investors pushing for a valuation below the $1.5 trillion mark discussed last week.
The new model scores 65 on the Artificial Analysis Intelligence Index but costs $26.54 per task and takes 44.6 minutes to complete, making it a premium option for complex, high-stakes workloads.
One short email when it matters. No recaps of recaps.