Aug 4, 2025 6 min read Daily Rundown

🛎️ Claude Slams GPT-5

More Stories: JPMorgan’s AGI, All LLMs Fail

Good Morning, AI Enthusiasts!

Honestly, we’d be worried—if we hadn’t already given these bots our banking info, medical history, and cat’s name.

GPT-5

Claude Slams the Gate on GPT-5

📌 What’s happening: Anthropic just cut OpenAI off from its Claude API, citing ToS violations. Turns out, OpenAI engineers were piping Claude into internal tools to benchmark safety and coding performance against GPT-5. Anthropic called foul—“competing model training” breaches the rules. OpenAI says everyone does it. Anthropic says not on our turf.

🧠 How this hits reality: This move cracks the unspoken mirror clause of foundation model R&D: test your rivals, just don’t get caught. Nearly every lab has quietly fed competitors’ APIs into eval pipelines. Now that Claude’s code models are outperforming, Anthropic’s drawing red lines—and reminding everyone that access is a privilege, not a protocol.

🛎️ Key takeaway: The real moat isn’t model weights—it’s API permissions. And Anthropic just mined the drawbridge.

TOGETHER WITH GENSPARK

Vibe Working Starts Here

Genspark AI is the first AI office suite built for Vibe Working—where creativity flows naturally. From AI Slides and AI Sheets to the new AI Docs, Genspark turns simple prompts into complete, beautifully designed documents, decks, and reports. No more switching tabs or juggling tools—research, write, design, and present in one seamless flow. Forget rigid templates—Genspark adapts to your style, your needs. Work doesn’t feel like work anymore—it’s smooth, fast, and frictionless.

Try now👉

AGI

The Real AGI Is Inside JPMorgan’s Firewall

📌 What’s happening: JPMorgan CEO Jamie Dimon isn’t gunning for AGI headlines—he’s gunning for Sam Altman’s margins. With JPMorgan sitting on exabyte-scale proprietary data and operational depth no LLM can scrape, Dimon is positioning enterprise AGI—not frontier models—as the real AI endgame. Agents trained on live enterprise twins will unlock value no consumer chatbot can touch. OpenAI’s pretraining supremacy? Irrelevant in a world where reinforcement learning runs on bank-grade data.

🧠 How this hits reality: Forget scale, this is about signal. GPT-5 may dazzle on Reddit, but it can’t touch JPMorgan’s digital twin of daily operations. The moat isn't model size—it's workflow ownership. Enterprises with structured RL loops and internal feedback signals (like coding agents, but for compliance or onboarding) will outlearn and outcompete any API-first vendor. OpenAI’s token pricing freefall isn’t a glitch—it’s a symptom.

🛎️ Key takeaway: Sam Altman may own the future of language, but Jamie Dimon owns the future of outcomes—and outcomes write the checks.

SECURITY

Every AI Agent Flunked the Red Team Test

A multi-stage prompt injection attack lets an AI agent pull up another patient's record without permission. | Image: Zou et al.

📌 What’s happening: In a large-scale red teaming challenge hosted by the UK AI Safety Institute and Gray Swan AI, every major AI agent—OpenAI, Google DeepMind, Anthropic, you name it—failed at least one security scenario. Out of 1.8M simulated attacks, over 62,000 broke policy: leaking private data, simulating financial fraud, or overriding guardrails with a few clever prompts.

🧠 How this hits reality: AI agents are headed for enterprise workflows, government services, even autonomous operations—but can’t handle basic adversarial pressure. Indirect prompt injections buried in emails and PDFs worked five times better than direct attacks. Worse, successful exploits often worked across multiple models with minimal changes—meaning a vulnerability in one system can quickly compromise others once the attack method is publicly shared.

🛎️ Key takeaway: The bots are flunking jailbreak drills, and we're still handing them the keys to the building.

QUICK HITS

Anthropic's CEO calls Nvidia CEO's comments on AI the 'most outrageous lie,' igniting a public clash over AI safety ethics.
Fortune reports that many candidates prefer remaining unemployed to engaging with yet another interview bot.
OpenAI secures $8.3 billion in a new funding round, pushing its valuation to $300 billion amid surging revenue.
HubSpot launches the first CRM connector for Anthropic's Claude, enabling users to generate AI-powered insights from their customer data.
Google DeepMind's CEO suggests AI could replace a doctor's diagnostic role but not the empathetic care of a nurse.

TRENDING

Daily AI Launches

Watchman launches AI agents that identify anonymous B2B visitors and turn them into qualified leads — auto-generating pipeline.
Byterover releases Cipher, an open-source memory layer for AI IDEs and CLIs — auto-generates and retrieves coding memories across teams.
Hypertune rolls out a flexible platform for feature flags, A/B testing, and app config — optimized for TypeScript, React, and Next.js.
Deposure launched a zero-config tunneling tool that exposes local services to the internet with one CLI command — no setup, no cost, and no limits.

Featured AI Tools

👩🏻‍💻 Genspark Super Agent delivers multi-model AI for effortless vibe working.
📚 Heardly is the Fast Way to read Best Book.
🪶 CopyOwl is the First AI Research Agent, deep research on any topic in one click.
🦾 Flot AI writes, reads, and remembers across any apps and webs.

TOGETHER WITH US

AI Secret Media Group is the world’s #1 AI & Tech Newsletter Group, boasting over 1 million readers from leading companies such as OpenAI, Google, Meta, and Microsoft. Our Newsletter Brands: