🔒 Anthropic's 154-page report exposes global Claude misuse 🔒
Anthropic's report reveals alarming misuse patterns challenging current AI regulation.

Anthropic released its first major threat intelligence report, a 154-page document detailing misuse it disrupted between December 2025 and August 2026. Among the most alarming cases: a scientist sought help with gain-of-function research on the chikungunya virus, someone in Yemen used Claude Code to build rocket guidance software, and a surveillance system was designed to monitor 25 million phone lines in Mali. The report also names seven Chinese AI labs—including DeepSeek, Alibaba, and Xiaomi—that used thousands of fake accounts to distill Claude's capabilities, with some serving Claude outputs to their own customers as proprietary models. Two macro trends stood out: AI-enabled cyberattacks are growing more autonomous, and sophisticated attacks no longer require sophisticated attackers—a critical signal for how regulation must evolve.

💰 OpenAI ships ChatGPT for Wall Street workflows 💰
GPT-6 Astra powers OpenAI's most aggressive vertical pricing play yet.

OpenAI unveiled ChatGPT for Financial Services, a specialized product powered by GPT-6 Astra with built-in data from Daloopa, PitchBook, LSEG News, and Crunchbase. Co-developed with Morgan Stanley and Evercore, it targets core finance workflows including research, financial modeling, pitchbook creation, and client materials. Integrations with S&P Capital IQ, MSCI, and Moody's let firms connect existing data subscriptions. Astra can parse financial statements, tables, and footnotes, then output Excel models, research notes, slide decks, or charts—all with traceable sources. This puts direct competitive pressure on Bloomberg Terminal and FactSet, signaling OpenAI's intent to own high-value enterprise verticals.

📱 Instinct raises at $10B as personal agents surge 📱
Instinct's $10B valuation signals a pricing reset for consumer AI agents.

Personal AI agents are rapidly becoming the next major consumer platform. Meta's Muse has rocketed to second place on app store charts, while xAI's Grok Bot has become one of the most discussed AI products in recent weeks. But the standout signal is Instinct, the invite-only agent app now reportedly raising capital at a $10 billion valuation—surpassing Snap's $9 billion market cap. This marks a clear adoption signal for persistent, personalized agents that act on users' behalf, moving well beyond general-purpose chatbots. The convergence of major players like Meta and xAI alongside ambitious startups suggests personal agents could define the next era of consumer AI distribution.

INTERESTING TO KNOW

🎙️ GPT-Live-1 hits API at $0.05 per minute 🎙️

OpenAI released GPT-Live-1 into its API at $0.05 per minute, a rollout that gives developers full-duplex speech capabilities—meaning the model can listen and speak simultaneously. Supporting 12 voice options, real-time interruption handling, and style control via system prompts, it showed nearly 80% fewer interruptions compared to turn-based systems. At roughly $72/day for continuous operation, this positions OpenAI as the pricing leader for production voice agents at scale.

🛠️ Cognition ships Devin Voice powered by SWE-2 🛠️

Cognition launched Devin Voice, enabling users to verbally instruct its AI software engineer, with a rollout powered by GPT-Live and the new SWE-2 model. Based on Kimi K3, SWE-2 achieves 50.0% on FrontierCode 1.1 Main1, matching GPT-5.6 Sol and Fable 5/5.1 at a fraction of their pricing while being 64% cheaper overall. This puts serious competitive pressure on frontier coding models by delivering near-top performance at roughly a quarter of GPT-6 Astra's cost.