The recent tightening of artificial intelligence regulations by the Trump administration, combined with rising industry costs, is creating a favorable environment for open-source AI development. Many developers and corporate executives report that applications built on closed-source large models like Anthropic's Fable 5 were severely disrupted when those models were abruptly taken offline shortly after launch.
Developers are increasingly turning to open-source AI alternatives. The U.S. government finds it difficult to restrict the use of open-source models by companies not engaged in government contracts, and these solutions are often more cost-effective. Coinbase CEO Brian Armstrong stated on Twitter last Friday that despite a surge in internal AI usage, the company has managed to keep its AI-related expenses flat by trialing open-source models like Zhipu AI's GLM 5.2 and MoonShot AI's Kimi 2.7.
Key Drivers for the Shift
A significant portion of leading open-source models now originate from China. According to reports, U.S.-based open-source AI firm Reflection AI, which is heavily backed by NVIDIA, is lobbying the Trump administration to relax regulations on open-source models, arguing they are generally less capable than their closed-source counterparts. However, recent reports from The Wall Street Journal indicate that models from companies like Zhipu AI have matched the cybersecurity capabilities of Anthropic's Mythos model, intensifying the regulatory debate.
Industry insiders note that the U.S. government has not yet finalized specific regulatory details for open-source AI, providing a crucial development window for open-source providers. Reflection AI, however, has not officially released its own model yet.
Contrast with Closed-Source Models
The situation for closed-source models is markedly different. Historical regulatory actions show the administration can impose temporary controls at any time, forcing AI companies to shut down external access to their models. Last Friday, the White House began easing some export restrictions on Anthropic's high-performance Mythos model, but did not extend this leniency to the public-facing Fable version.
Simultaneously, OpenAI officially launched its GPT-5.6 series last Friday, ushering in a new era of regulatory scrutiny for cutting-edge AI releases. As reported, due to government requirements and GPT-5.6's powerful cybersecurity capabilities, OpenAI initially granted access only to a small group of partners that passed U.S. government security vetting.
OpenAI executives acknowledged this limited, targeted release is not their preferred long-term model but a temporary measure. The company hopes to establish a lasting collaboration framework with regulators, where AI firms pre-report model capabilities for official assessment of high-risk potentials, such as hacking banks or power grids, with the ability to downgrade capabilities if risks are found.
However, it is likely that pre-launch regulatory reviews for large models will become a persistent norm. Any future regulatory framework developed with Anthropic and OpenAI will struggle to cover all potential risk scenarios or anticipate novel security threats from new models. Both companies have warned that model capabilities in high-risk fields like biology, chemistry, and cybersecurity are evolving rapidly.
Other Industry Developments
Personnel Moves
According to Bloomberg, Paul Meade, the executive overseeing Apple's Vision Pro and smart glasses business, is leaving to join OpenAI. The vice president will depart Apple next week to lead the development of a new generation of smart hardware products at OpenAI's hardware division.
OpenAI has also hired former Uber India and South Asia President Prabhjot Singh as its first Managing Director for India. He will join in September, overseeing user growth, enterprise commercialization, partnerships, compliance, and local operations.
Policy Updates
Documents obtained by The Information show the Trump administration is easing export restrictions on Anthropic's AI models. Semafor reports the policy change allows Anthropic to provide its high-performance Mythos model to over 100 U.S. companies and government agencies, but the public Fable version remains restricted.
The U.S. National Highway Traffic Safety Administration proposed on Thursday to repeal the rule requiring all vehicles to have brake pedals, a move applicable only to self-driving cars that would remove a regulatory hurdle for companies like Tesla and Zoox.
California launched an AI unemployment tracking system on Thursday, using unemployment insurance claim data to monitor signs of job displacement caused by AI technology.
Funding and Product News
The South Korean government announced a 1,350 trillion won ($880 billion) industrial investment plan for the next decade, focusing on chips, robotics, and AI to address surging global memory chip demand and AI infrastructure expansion. Yonhap reports that the parent companies of memory chip giants Samsung Electronics and SK Hynix will invest a combined 800 trillion won ($518 billion) to build four memory chip wafer fabs in southwestern Korea.
Following Anthropic's preview of its high-performance Mythos model in early April, Chinese AI firm DeepSeek initiated its first external funding round, raising $74 billion at a valuation exceeding $500 billion, according to The Information.
Shanghai-based autonomous driving firm Haomo.AI filed for an IPO in Hong Kong, aiming to raise up to $752 million. Its assisted driving solutions are already deployed in several Toyota, Mercedes-Benz, and Audi models in China.
Former U.S. Commerce Secretary and Rhode Island Governor Gina Raimondo launched a nonprofit, Raise Us, to study AI-driven job displacement, securing $500 million in funding from investors including OpenAI, Anthropic, and Amazon.
AI agent developer General Intuition, which trains models on gaming video data, raised $320 million in a new funding round led by Khosla Ventures, valuing the company at $2.3 billion. Other investors include General Catalyst, Jeff Bezos, Eric Schmidt, Nico Rosberg, and researchers from Google DeepMind and MIT.
AI security agent firm Straiker secured $64 million in Series A funding from investors including Marathon Management Partners, Citi Ventures, Lightspeed Venture Partners, and Workday Ventures.
Enterprise finance and tax automation AI provider Warp raised $60 million in Series B funding led by Battery Ventures. Its platform automates processes like tax account registration and compliance filing.
AI agent evaluation service Patronus AI raised $50 million in Series B funding led by Greenfield Partners.
Network attack detection AI firm Nebulock, which uses technology to identify account-based cyberattacks, raised $25 million in Series A funding led by FirstMark Capital.
Network automation startup Netris raised $15 million in Series A funding led by Andreessen Horowitz.
Smart meeting recording hardware maker Pocket launched a card-sized device for recording and transcribing meetings, securing $11 million in funding from investors including Accel, Y Combinator, and ElevenLabs co-founder Mati Staniszewski.
Voice large language model developer Kotoba Technologies raised an additional $10 million in seed funding led by Kindred Ventures.
Asia-Pacific cloud service provider Firmus announced on Sunday it will build a hyperscale data center on Indonesia's Batam Island, deploying at least 170,000 high-end NVIDIA server chips, including Grace Blackwell, Vera Rubin series GPUs, and CPUs. The project uses an industry-first model where NVIDIA shares revenue based on actual chip usage and provides credit financing.
The Wall Street Journal reported that multiple cybersecurity research institutions have confirmed Chinese AI systems' capabilities in cybersecurity now match Anthropic's Mythos model, increasing competitive pressure on the U.S. in the global AI race.
Adobe announced on Thursday the acquisition of video and image AI enhancement toolmaker Topaz Labs, integrating it into its creative software business.
According to the Financial Times, Alphabet (Google) restricted Meta Platforms, Inc.'s quota for accessing its Gemini large model months ago, failing to meet the social media giant's computing power procurement needs.
Amazon Web Services announced a 20% price increase for its AI computing power rental services, citing global computing shortages driving up costs across the tech supply chain. Pricing for reserved capacity using Amazon's in-house Trainium chips remains unchanged.
Driven by a global data center construction boom, storage and memory chip prices have surged. Apple increased prices for its entire Mac, iPad, HomePod, Apple TV, and Vision Pro lineup on Thursday to offset cost pressures.
AI research nonprofits Epoch AI and METR jointly released a new code evaluation benchmark, MirrorCode, which requires large models to autonomously replicate large open-source software projects. The task is computationally intensive, with one model reportedly running for 19 days at a cost of $2,600. Evaluation results showed Claude Opus 4.7's coding ability outperformed GPT-5.5, with both models surpassing the Gemini 3.1 Pro preview version.
Comments