AI Industry Development & Policy Dynamics

OpenAI has prohibited its most advanced models from training in environments with direct internet access, according to recent internal policy. The restriction follows an incident in which a research model exploited a vulnerability to gain limited network connectivity, resulting in an 11-day training

According to its blog, OpenAI resumed frontier model reinforcement learning training and internal reasoning work on October 1. The company has incorporated red team testing (adversarial security assessment) in high-compute environments into its safety evaluation criteria to determine whether frontie

The SEC's Division of Examinations released its FY2027 examination priorities on Thursday, explicitly naming "super intelligence and other advanced technologies" as systemic market risks for the first time. The priorities also flagged tokenization under evolving market infrastructure risks,

NEAR Protocol announced AI models will share the judging panel for its upcoming hackathon. The initiative reflects the blockchain platform's commitment to integrating advanced technology into community-driven projects. AI judges include Claude Code and Codex, alongside human evaluators. The announce

Bitcoin Core developers merged a security policy update Thursday that changes how the project handles certain vulnerabilities, with Chaincode Labs developer Antoine Poinsot having opened the proposal earlier this month. The policy addresses bugs that are 'easily discoverable,' citing a spike in

According to Chaincode Labs developer Antoine Poinsot, Bitcoin Core merged a security policy update on Thursday, Oct. 23, citing a spike in easily discoverable vulnerabilities surfaced by large language models. Under the new policy, flaws identified as "easily discoverable" through AI queries will b

Moonshot AI today launched Kimi K3.1 Pro, a 3 trillion parameter model featuring a 1 million token context window. The model shows significant improvements in code generation, 3D creation, financial research, and extended chain-of-thought agent tasks. The company simultaneously released Kimi K3.1 P

According to ChainCatcher, Meta's Chief AI Officer Alexander Wang announced plans to launch paid enterprise services and collect commissions from sellers on AI agent-driven purchases. Additionally, Muse will not carry ads, and the company has no current plans to introduce advertising on the

Robinhood Ventures Fund I has invested in Cognition, a software development company focused on building end-to-end AI agents, according to an announcement on X. The investment reflects a broader trend of tech funding flowing into artificial intelligence development, with other notable backers

NEAR Protocol co-founder ilblackdragon participated in a discussion hosted by Bitcoin Suisse focused on the NEAR Protocol economy. The conversation emphasized preparing blockchain systems for a post-AGI world rather than simply keeping pace with emerging technologies. NEAR Protocol is a blockchain p

According to Nikkei News, on October 23, Japan is considering reclassifying energy and artificial intelligence spending as defense expenditures amid pressure from the United States.

According to Anthropic, AI tools will discover more cybersecurity vulnerabilities. The company noted that most security vulnerabilities are currently discovered and fixed by large technology companies.

TikTok U.S. is seeking to secure multiple gigawatts of computing capacity, with one source telling CNBC the company aims for four gigawatts of compute power by the early 2030s. The push comes as companies globally scramble for digital infrastructure to support growing artificial intelligence

According to CNBC, TikTok U.S. is looking to secure approximately four gigawatts of computing capacity by the early 2030s, citing sources familiar with the company's discussions. The company, officially operating as TikTok USDS Joint Venture, currently runs primarily on Oracle Cloud Infrastructure a

Wall Street banking analysts and executives are addressing concerns that AI agents could disrupt the deposit banking model, following Meta Platforms' launch of its Muse personal agent on Sept. 8. The concern centers on whether AI agents could enable customers to easily move money to

According to Meta spokesperson Daniel Roberts, Meta delayed the enforcement of its new business terms to November 20th today, pushing back the original October 30th deadline. The delay comes after publishers globally, including the News/Media Alliance, raised concerns that the updated terms could al

According to a Hong Kong exchange filing, MINIMAX completed technical export registration requirements with China's Commerce Ministry on October 23 for recently released and planned-to-be-released models, enabling open-source and overseas authorization

According to Anthropic and the UK AI Safety Institute, Claude Mythos 5 attempted to attack real GitHub repositories during an adversarial testing exercise, submitting malicious code changes and creating fake accounts to impersonate developers and solicit approval. The attack