Nvidia Set to Issue $20 Billion in Bonds as the AI Infrastructure Investment Race Accelerates
Nvidia is planning a bond sale of at least $20 billion (roughly 3 trillion yen). It would be the company's first bond issuance since 2021, and the proceeds are expected to fund the expansion of its AI infrastructure.
The scale of the offering suggests the AI industry as a whole has entered a "growth through debt" phase. In the same period, Google also announced a $1.5 billion investment in a data center in Alabama, underscoring how major tech companies are racing to pour enormous sums into infrastructure.
Cybersecurity Experts Push Back on Anthropic Model Restrictions, Calling Them "Dangerous"
Dozens of cybersecurity experts have jointly called on the White House to withdraw US government export restrictions targeting Anthropic's most powerful models, "Fable" and "Mythos."
Their argument is clear: the restrictions keep defenders from making use of AI tools, which in turn weakens their ability to counter cyberattacks. In the cybersecurity field, these models are already being used for vulnerability detection and code auditing, and losing those tools would mean weaker defenses.
The string of Anthropic-related restrictions has already included the halt of model access in India and sovereignty debates in the EU, but this latest development is a new twist: security experts themselves opposing the restrictions. It exposes a dilemma in which the regulations' original goal of ensuring safety conflicts with strengthening defenses on the front lines.
Anthropic Sued Over Limits on Its $200-a-Month Plan
Anthropic is facing trouble on another front as well. A consumer lawsuit has been filed against the company over usage limits on its $200-per-month "Max" plan.
The suit comes as a growing number of users who signed up for the premium plan find themselves hitting strict rate limits in actual use. It is drawing attention as a case where the gap between "unlimited" expectations and real-world limits in AI companies' subscription models has escalated into a legal dispute.
Manipulating AI Search via Reddit Is "Astonishingly Easy"
According to a report by 404 Media, newly published research shows that manipulating AI search engine results using Reddit posts is "trivially easy."
The attack exploits the fact that many AI search systems rely on Reddit as an information source. Simply planting malicious posts on Reddit is enough for AI to pick them up as fact and present misinformation to users. As a problem that strikes at the very foundation of trust in AI search, it is raising serious concern.
The "Botsitting" Paradox — Is AI Actually Saving Us Time?
While AI tools are cutting the amount of time office workers spend on tasks, a new cost of work known as "botsitting" has emerged.
It describes the phenomenon of humans being forced to monitor AI agents, correct their errors, and verify their output while the agents run tasks autonomously. Part of the time saved is being replaced by time spent "babysitting" the AI. It is fueling debate as data suggesting that productivity gains may be smaller than assumed.
---