Lectura

Darknet sellers are offering large-model accounts at as little as 3% of the official price, and enterprises' self-hosted AI compute has been named a new target.

According to the Financial Times on September 26, John Hultquist, chief analyst at Google's threat intelligence team, said "LLM hijacking" has risen sharply this year: darknet markets are selling access to models from Anthropic, Google and OpenAI at discounts of up to 97%, with some sellers promising free replacement credentials if an account is banned.

Attackers use this cheap access to expensive models for ransomware, cyberwar and espionage; defenders rely on the same tools, which widens the cost gap.

Hultquist warned that as more companies deploy custom models on their own servers, that self-paid compute is becoming a resource attackers covet and should be defended as a high-value target.

Voice AI startup Modulate announced on September 28 that it raised $25 million in new funding, led by Future Ventures with returning investors Hyperplane and Lakestar participating, bringing total funding to $60 million. The announcement did not disclose a valuation.

Modulate's models work on the raw audio of a conversation, reading emotion, tone, intent and signs of a deepfake voice. They are used for game voice moderation, screening callers impersonating hospital staff, and checking how voice AI agents perform. The company says its platform now handles more than 10 million hours of audio a month.

The new money goes toward SDKs and APIs for developers and models built for specific industries. The round is a company announcement as reported by SiliconANGLE; the figures and stated use of funds are the company's own.

Bill Gates believes humanity must get through an adjustment period of roughly 20 years before AI delivers broad benefits.

On September 27 local time, Microsoft co-founder Bill Gates told NBC's Meet the Press that AI development has two phases: today's AI has the potential to tackle high living costs and expensive healthcare but cannot truly solve them yet; once AI combines with humanoid robots, building houses and growing food will end mass shortages — an era he calls one of abundance.

He also warned that AI is now powerful enough to "drive an event that kills a billion people," and that law enforcement and political leaders must help define safety safeguards and monitoring mechanisms, which should become mandatory. These are his statements in the interview; a full transcript was not published with the report.

Neural networks are the kind of programs behind chatbots and image recognition, internally made of massive numbers of mutually weighted digits, with no one having written the rules line by line. In a long interview, Chris Olah discussed the direction he is betting on: mechanistic interpretability—opening up the network, reading out the steps it actually performs, and giving an account of each step that can be tested and potentially overturned.

Today, when AI companies talk about safety, they often say they want to "open up the model and look inside," following exactly this program. The objection he himself worries about most in the interview is that as model scale grows, this kind of disassembly may not keep up, and this remains unresolved to this day. When you see claims like "we understand our own models," first ask: how large a model did they actually open up.

The words in the interview are expectations about a research direction, not verified conclusions. If someone uses it to assert that some large model has already been understood, or can never be understood, don't use it to judge—that goes beyond what the interview can support.

"Chris Olah on what is going on inside neural networks" (2021) | Next review 2027-09-20