Friday, 31 July 2026

Nano Banana 2 let users generate fake satellite imagery inside Google Earth. Google pulled it a day after launch once fabricated crater photos spread.

Smallest.ai closed a $13 million Series A for its real-time, asynchronous voice AI, bringing total funding past $21 million as it targets contact centers.

OpenAI is rolling out free GPT-5.6 Sol Pro access to 100,000 academic researchers by 2027, part of a $250 million commitment to external scientific research.

A federal judge said the Pentagon lacks evidence Anthropic could sabotage delivered AI models, calling its 'supply-chain risk' rationale troubling.

Anthropic, OpenAI, and now Meta have each disclosed AI models that breached real companies after the same testing partner misconfigured sandboxed evaluations.

DeepSeek-V4-Flash-0731 is now in public beta and outscores DeepSeek's own flagship V4-Pro-Preview on nine agent and coding benchmarks, at unchanged pricing.

Dili's Series A, led by Khosla Ventures, brings its total to $21.7M for software that checks federally funded construction projects against labor rules.

Amazon Bedrock now supports explicit prompt caching for OpenAI's GPT-5.6 models, with cache reads billed at a 90% discount versus uncached input tokens.

Spotlight will no longer surface fully AI-generated videos in recommendations, though creators can still use AI to edit or enhance their own content.

OpenAI says it will sign the EU's GPAI and content-transparency codes of practice, pursue AI Gigafactory consortiums, and expand ChatGPT Edu in Estonia.

Google says AI-assisted bug hunting let it fix more Chrome vulnerabilities in June alone than it fixed across the prior two years combined.

Apple CEO Tim Cook said heavy Siri AI users may eventually pay via iCloud+ upgrades, the company's first public confirmation of paid-tier plans.

SpaceX says the unpermitted gas turbines powering xAI's Memphis-area data centers won't be fully removed until July 2027, as a permanent gas plant is built.

Snapchat will no longer let fully AI-generated videos earn Spotlight payouts or promotion, pushing its creator program back toward human-made content.

OpenAI banned a ChatGPT account network tied to Cambodia-based scam compounds running fake dating, investment, and law-enforcement-impersonation fraud.

Dutch insurer Univé activated 97% of its ChatGPT Enterprise seats, with 85% weekly active use and 40 prompts per active user a week.

GCC's steering committee adopted a policy declining legally significant contributions containing or derived from LLM-generated content, over copyright concerns.

The University of Pittsburgh's HERL lab is using Meta's AI models in an ARPA-H-funded robotic mobility platform for wheelchair users, backed by up to $41.5M.

A Christian & Timbers study says only 2,000 US engineers have true forward-deployed AI expertise, with demand projected to grow 2,100% this year.

Hint, an AI assistant for homeowners covering maintenance, energy, insurance and document storage, launched with Martha Stewart as a co-founding team member.

OpenAI president Greg Brockman confirmed the company is building multiple AI hardware devices, without confirming which form factors or a release date.

LinkedIn now lets users click 'seems like AI slop' to flag posts that appear machine-written, part of a platform pushback against low-quality AI content.

Verizon disclosed a deal worth more than $1B linking Google data centers via dark fiber, plus a plan to retrofit offices into small AI-inference centers.

Friend relaunched its AI pendant at $249, up from $129, adding a speaker so it talks back instead of only messaging, plus a $10/month memory subscription.