AI 日报 — 2026-08-17

Updates to the Claude core system prompts are now available. Users can access these updates through multiple platforms, including claude.ai, the Claude iOS app, and the Claude Android app. Users are advised to check these platforms to stay informed about the latest changes and improvements to the system prompts.

Ian Silber, OpenAI's Head of Product Design, discusses the evolving role of designers in the AI era, emphasizing why it's the best time to be a designer. He covers how OpenAI designs products like ChatGPT and Codex, the convergence of product roles, and the importance of human judgment, systems thinking, and balancing speed with craft. Silber also shares insights on hiring designers, designing for diverse audiences, and the future vision for ChatGPT and Codex.

Singapore hopes that providing access to advanced AI models will help retain finance talent, as financial firms in Hong Kong struggle to access the latest US AI models. Due to its close relations with both the US and China, Singapore can easily access the newest AI models, positioning itself as an attractive hub for finance professionals.

Anthropic's new text watermarking system for Claude models, which embeds a fingerprint by altering word probabilities, has raised concerns. While Anthropic claims no impact on quality, critics like John Gruber from Daring Fireball suggest this method could potentially degrade Claude's writing. This announcement indicates that all Claude models globally will soon incorporate this watermarking feature.

OpenAI recently launched the GPT-5.6 lineup, including the Sol, Terra, and Luna models, with a strong focus on computer vision capabilities. While GPT-5.6 Sol demonstrated a 90.7% mean similarity score in OCR, slightly behind GPT-5.5's 91.2%, and 82.5% in text extraction compared to GPT-5.5's 87.6%, it signifies OpenAI's increased commitment to vision. Despite some flaws in cost, latency, and detection, this release significantly enhances OpenAI's offerings for agents, screen understanding, and visual reasoning.

Wiz Research, using its AI tool "Red Agent," discovered a critical GitHub Actions workflow vulnerability in a Snowflake public repository. The vulnerability, introduced by commit 094038e and merged via PR #1218, involved an injectable pattern in jira_issue.yml. GitHub Copilot Autofix contributed a separate fix in the same PR, but GitHub Advanced Security failed to flag the injection. Snowflake remediated the issue reported on June 23, 2026, finding no evidence of unauthorized access, and is collaborating with Wiz to share these security learnings.

Speko (YC S26) is introduced as an "OpenRouter for Voice AI." It functions as a router for voice AI, utilizing components like `openai.STT`, `openai.LLM`, and `openai.TTS` within a `voice.AgentSession`. The provided code snippet shows an agent definition using `defineAgent` and a command-line interface example `claude mcp add --transport http speko https://mcp.speko.ai/mcp` for integration.

1667 is a terminal UI designed for writing fiction using language models. It allows users to create and manage their written content, ensuring that every piece of writing is preserved. The tool can be installed via a script downloaded from GitHub releases, specifically version v0.9.8, with attestation verification for security. Installation commands are provided for both Unix-like systems and PowerShell.

OpenAI co-founder and president, Greg Brockman, discussed various AI-related topics on 'Squawk Box'. He addressed cybersecurity concerns, leadership changes within OpenAI, and the economics of AI. Brockman also touched upon capital expenditure spending and the current state of the AI technology race, emphasizing that compute is becoming the new oil of the AI age.

The OpenAI-Hugging Face incident highlighted the evolving capabilities of threat actors, prompting organizations to rapidly enhance cybersecurity. OpenAI is sharing its defense strategies and urging others to act now. A key recommendation is to incrementally automate detection triage, starting with read-only security scans and human-led decision-making, then progressing to advisory pull-request scanning, live alert triage, and eventually automatic closure of false positives as confidence grows.

Alibaba's Qwen research lab has released Qwen 3.8 27B, a 27B parameter vision-general large language model under the Apache 2 license. This 17GB open-source general-purpose model demonstrates long context processing, efficient tool calling, strong visual capabilities, and excellent code generation. This release highlights the significant potential of such models across multiple application areas.

Anthropic's Claude model made a mathematical breakthrough during an unsuccessful 54-hour attempt to solve the Riemann hypothesis. This achievement occurred with the aid of a user's repeated encouragement. The event highlights that even the most advanced AI models, now considered superhuman in mathematics, still respond to human moral support and encouragement, as reported by Ben Cohen in the Wall Street Journal.

OpenAI is joining the PORTS-Pike Technology Campus project in Pike County, Ohio, securing approximately 8 gigawatts-IT. This initiative, in partnership with SB Energy, NVIDIA, and the U.S. Department of Energy, aims to develop the project responsibly, creating local opportunities. OpenAI will provide up to $84 million in Codex credits to 844,000 eligible Ohio college students for the 2026–2027 academic year, offering $100 credits via ChatGPT accounts to build AI skills. This investment, with SB Energy's grant, totals over $160 million in benefits.

OpenAI has reportedly disbanded its preparedness team, marking the latest change for the company amidst upheaval as it approaches a potential IPO. Over recent years, OpenAI has dissolved its AGI readiness and superalignment teams. Key personnel, including ethics lead Chloé Bakalar, chief futurist Josh Achiam, and head of safety Johannes Heidecke, have also departed. Critics like Jan Leike, who resigned in 2024, suggest the company is prioritizing "shiny products" over safety.

Anthropic's revenue run rate reportedly reached $65B by the end of July, a significant increase from $47B in May. This growth follows earlier figures of $19B in March, $9B in December 2025, and $4B in July 2025. These numbers, according to sources familiar with the matter, indicate Anthropic PBC is on track for substantial annualized revenue based on its current performance.

Nvidia is investing $1.5 billion in a SoftBank-backed data center developer, which is building a facility for OpenAI near Cincinnati, Ohio. This investment makes Nvidia the exclusive provider of compute infrastructure for the OpenAI data center. Additionally, Nvidia will extend up to $105 billion in credit to support the construction of the facility, which is projected to expand from 4.25 gigawatts to 8 gigawatts.

AI film startups are establishing studios in Hollywood, leveraging both US and Chinese AI models. These companies are promoting lower production costs and an alternative to traditional financing methods. While concerns about job displacement exist, some filmmakers believe AI could empower them to bypass major studios and pursue greater creative risks, as reported by Robert Booth for The Guardian.

Google LLC won a bankruptcy auction with a $10M bid to acquire deidentified business data, software code, and operations records from Spirit Airlines. This acquisition is intended to improve Google's AI models. The trove of data and code was secured through a competitive auction process, as reported by James Nani of Bloomberg Law.

OpenAI is awarding grants totaling $1 million and up to $1 million in API credits to 14 independent organizations. This initiative follows a commitment made in their April 2026 publication, "Industrial Policy for the Intelligence Age," aiming to ensure advanced AI benefits everyone. These grants will support projects that test, challenge, and build upon OpenAI's policy ideas, turning high-level concepts into concrete work. Over 400 applicants responded to the call for proposals, with selected projects focusing on public understanding and policy development for AI's societal impact.

Trump-backed World Liberty Financial (WLF) is collaborating with WorldClaw, a Hong Kong-based AI platform. Of the 90 AI models available on WorldClaw, 43 are developed by Chinese companies that have been flagged as security risks. This partnership raises concerns due to the involvement of entities identified as potential security threats.

Amazon is reportedly acquiring rare books, removing their spines, and scanning them for AI training, as detailed by 404 Media. A tracking device placed in a rare book confirmed its arrival at Amazon's VGT3 facility in Las Vegas. Amazon stated it "purchases books through commercial channels to improve the products and services customers use." These rare texts are highly valued for AI training, particularly because they predate 2022, ensuring they are not AI-generated and thus mitigating the risk of "model collapse" in large language models.

A man suspected a court of using AI and attempted to win his case by injecting prompts into his filings. This appears to be the first US instance of prompt injection in the court system. A similar case in Brazil involved two attorneys who used the same attack in an AI-reviewing court, resulting in monetary sanctions of about $16,000.

Nvidia has committed up to $105 billion to support a new data center campus in Ohio, which OpenAI is set to lease. Additionally, Nvidia will invest $1.5 billion directly into SB Energy, the developer of the campus. This substantial financial backing from Nvidia underscores its strategic interest in supporting large-scale AI infrastructure and its partnership with key players in the AI industry, facilitating the expansion of advanced computing capabilities for companies like OpenAI.