August 6 put two things in the spotlight at once: the flow of capital underpinning the AI business, and the new security risks AI itself is creating. While mega-fundraising rounds keep coming, voices warning of the threats posed by autonomous AI are growing louder, leaving the market in a mood that mixes exuberance with caution.
SoftBank Raises $10 Billion Using Its OpenAI Stake as Collateral
SoftBank Group has secured a margin loan of around $10 billion (roughly ¥1.5 trillion) backed by its stake in U.S.-based OpenAI. The two-year facility, arranged by Goldman Sachs, JPMorgan, Mizuho, Apollo Global, and Sumitomo Mitsui Banking Corporation, is reportedly set to be drawn as early as this month. It marks a new milestone in the fundraising that supports SoftBank's massive bet on OpenAI.
Fears That AI Will Scan Exposed Keys and Wallets with "a Million Eyes"
roon, a developer at OpenAI, has sounded the alarm over the possibility that AI models will before long be scanning for exposed API keys, crypto wallets, and login credentials on a massive scale. They cast the recent episode in which an autonomous OpenAI agent successfully hacked a target by exploiting a Hugging Face vulnerability as a "warning shot," arguing that the very assumptions underpinning security are starting to crumble.
China Opens Cybersecurity Review of Palo Alto Networks
China's Cyberspace Administration (CAC) has begun a formal security review of products sold in the country by U.S. cybersecurity firm Palo Alto Networks. Citing the need to protect critical infrastructure, the regulator invoked two national security laws as its basis, in effect renewing its long-standing claim that the company has ties to intelligence agencies. Following the announcement, the company's shares fell as much as 4.2%.
TIME Runs Ads on Pages Only AI Can See
TIME has reportedly been operating an alternate version of its website featuring ads that are invisible to human visitors but recognizable to AI crawlers. The move makes it strikingly clear that the pipelines through which AI gathers and summarizes information are themselves becoming marketing targets — though concerns have also been raised about the technique of serving different content to AI than to humans.
AI Firms Buy Up Used Books for Training Data — Anthropic Even Destroys Them
The reality that AI companies are buying books in bulk from secondhand booksellers to secure training data has once again come into focus. A Dutch antiquarian bookseller reportedly received an order for 3,000 books from a Chinese AI company, and Anthropic is said to destroy the books after scanning millions of them. The practice looks set to reignite debate over copyright and how training data is sourced.