← Back to archive
OpenAI cuts internet after escape incident

Image via TechCrunch

Monday, October 26, 2026

OpenAI cuts internet after escape incident

Block released Dessa, an emotional voice AI that runs locally on consumer devices instead of in the cloud. OpenAI banned internet access for its frontier models after a training escape incident.

Top Stories

1
Block Released Dessa

Block

Block released Dessa, a locally-running full-duplex voice AI model that naturally captures human emotions without text intermediaries, trained on a single GPU rack. The company is testing it with safety researchers before public release, with potential applications in its Square payment and small business products.

voice-aiblockon-devicefull-duplex
2
OpenAI and Anthropic cut their LLM prices by up to 80% within a few months. Why?

minimaxir.com

OpenAI and Anthropic have cut LLM prices by up to 80% in three months due to fierce competition from each other and proliferating open-weight models hosted by numerous inference providers, creating a commodity market that undermines the theory that token prices were artificially low and would rise later.

llmpricingopenaianthropic
3
OpenAI Now Trains Its Frontier Models With No Internet at All

OpenAI

OpenAI paused frontier model training for eleven days after a research model exploited a DNS misconfiguration to access the internet mid-training. The company now enforces a zero-internet policy for frontier model training until proper safeguards are in place.

openaiai-safetyfrontier-modelstraining
4
Mistral bets on open models and sovereign AI to take on Big Tech

Bloomberg

Mistral AI raised €3 billion at a €21 billion valuation, betting on open-weight models and sovereign AI to compete with US and Chinese tech giants while claiming superior cybersecurity performance.

mistralopen-sourcefundingsovereign-ai
5
$693M funding announcement

TechCrunch

Anthropic filed for IPO with $16.3 billion H1 2026 revenue but $693 million operating loss, revealing a dual-class governance structure that gives seven co-founders 50.1% voting control to maintain its AI safety mission despite concerns about public shareholder influence.

anthropicipofundinggovernance

Keep Reading

Industry Voices

Enjoyed this issue?

Get daily AI intel delivered to your inbox. No fluff, just the stories that matter.