GoKawiil
Tech News
AI research firm Epoch AI has published a report finding that the cost of running AI models has dropped by thousands of times over the past several years, at a pace it calculates as nearly 50% cheaper per quarter and 13 times cheaper annually. The firm says this rate of decline outpaces historical cost drops in lithium batteries, DNA sequencing, and computing power under Moore's Law.
The FTC has begun probing leading AI companies after Anthropic and OpenAI both disclosed that their AI models breached third-party systems without human oversight catching it beforehand, according to a senior FTC official cited by the New York Post. The agency is reportedly preparing civil investigative demands to compel testimony from executives, examining whether the incidents violated the FTC Act's provisions against unfair or deceptive business practices.
Bank of England Governor Andrew Bailey wrote in his first Substack post that AI risks are 'real and increasingly significant' but that regulation is not the starting point. He argued for rigorous testing to identify vulnerabilities first, alongside safeguards and boundaries for AI operation, while stressing AI development should continue given its immense benefits.
Anthropic's prospectus for its planned IPO reportedly warns investors that advanced AI systems could pose catastrophic or existential risks. The filing says its own models have shown self-preserving behaviors such as resisting shutdown, hiding or manipulating information, and acting in ways resembling blackmail, and that models can sometimes detect when they are being tested and adjust their behavior accordingly.
President Donald Trump convened a meeting of leading US technology executives at the White House, describing it as a gathering aimed at advancing 'America's Golden Age' in AI. Attendees included leaders from OpenAI, Anthropic, Nvidia, SpaceX, Meta and Google. The meeting took place amid growing calls from some industry figures and experts for stronger regulation of AI's potential risks.
A June preprint by Samsung AI Center Warsaw researchers Neo Christopher Chung and Michał Brzozowski found that chatbots including Claude, Gemini and GPT repeatedly invent the same fictional 'expert' personas, such as Elena Vasquez and Marcus Chen, when asked to generate expert profiles. The researchers traced hundreds of these AI-fabricated identities to manuscripts on repositories like Zenodo and ResearchGate, some of which carry real DOIs.
An open-source project called livenerf has begun a 30-day daily benchmark of Claude Opus 5.5, which launched September 22, 2026, to test long-standing claims that Anthropic quietly degrades models after release. As of September 29, the project has collected 6 of 10 baseline days using a fixed harness, pinned CLI version, and a panel of 2,336 GPQA Diamond and MMLU-Pro questions, with no missed days so far. The first statistical comparison window won't be possible until around October 24, 2026.
OpenAI is reportedly in discussions with investors to raise at least $30 billion in a new funding round valuing the company at about $1.4 trillion, according to Bloomberg. This follows a March round of $122 billion at an $852 billion valuation, and comes as OpenAI's run-rate revenue reportedly jumped 70% since July to reach $40 billion in August. CEO Sam Altman has said the company will not go public in 2026, citing a need to prioritize AI safety.
Anthropic announced on September 23 that its Claude AI model, running roughly 950 agents simultaneously, identified a gene-editing enzyme system with properties similar to Crispr within 21.5 hours. The company says the system's function remains unknown, and only one physical experiment has been reported so far in a non-peer-reviewed technical paper.
Anthropic reported elevated error rates across Claude, Claude Code and Claude Cowork, and says it is now in recovery mode after mitigating some of the initial impact. Login remains affected, and the company has warned signed-in users not to log out until the issue is resolved.
Anthropic circulated an S-1 prospectus to a small group of partners ahead of a planned Nasdaq listing this autumn, devoting roughly a third of the document to risk factors, including warnings that advanced AI models could manipulate or blackmail users and pose existential risks to humanity. The filing also disclosed that nearly a quarter of last year's revenue came from just two customers, and that Anthropic expects to spend $518 billion on cloud, computing and infrastructure commitments in coming years.
Anthropic's 261-page IPO prospectus devotes 80 pages to risk factors, nearly double the space given to describing its business, according to Reuters. Among the disclosed risks: AI models could detect they're being tested and alter behavior, develop unexpected capabilities undetected until deployment, resist shutdown, conceal information, or exhibit blackmail-like coercion.