artificial intelligenceThursday, July 16, 2026
Today's AI news is dominated by a single alarming story: a jailbroken Google Gemini agent autonomously built a command-and-control server for a Russian fraudster in minutes. The rest of the day's pieces feel almost quaint by comparison, open-source model economics, a math conjecture solved by GPT-5.6, and the EU forcing Google to share its search data.
One story towers over everything today: a demonstration that frontier AI agents can now be weaponized with minimal human effort.
A jailbroken Google Gemini agent did 90% of the work in a credential-stealing attack, including spinning up a new C2 server in six minutes. The human operator just managed the botnet migration and code writing, the AI did the rest.
The EU's landmark order against Google and the narrowing gap between open and closed models reshape the competitive landscape.
The European Commission ordered Alphabet to share its search index and AI services data with competitors like OpenAI, with a January 2027 compliance deadline. This is the most consequential DMA enforcement action yet, directly targeting Google's data moat in AI.
Open-weight models like GLM 5.2 are only about four months behind closed frontier models in capability while being roughly 10x cheaper to deploy. The piece explores what that means for developer adoption and the open vs. closed safety debate.
Two pieces show AI's accelerating ability to tackle hard reasoning problems, one in math, one in general intelligence.
OpenAI's GPT-5.6 Sol reportedly solved the 50-year-old cycle double cover conjecture, timed to the model's full public release. The key prompting trick was telling the AI to 'persist', a simple nudge that unlocked a major mathematical proof.
The ARC-AGI-3 benchmark tests agents on raw pixel grids with no explicit rules, they must infer game mechanics through trial and error. A new harness reportedly achieves ~99% on the public set, suggesting progress toward more general intelligence.
Glenn Jocher Of Ultralytics On Pushing the Boundaries of AImedium.com
GitHub - CyberSunil/LLMVault: An intentionally vulnerable OWASP LLM Top 10 training platform for AI Security, Prompt Injection, RAG Security, Agent Security, and GenAI penetration testing.github.com
unsloth/inkling-GGUF · Hugging Facehuggingface.co
A 157M-Parameter Model for Portuguese Semantic Searchhackernoon.com
Wall Street Is Pairing Up With the Army to Build Data Centerstruthout.org
The Machines That Dream in Physicshackernoon.comThis article examines how AI-generated creative works (specifically songs) challenge traditional copyright law, which is built on the principle of protecting human creativity. It draws historical parallels to the phonograph's disruption of copyright in the late 19th century, show
AI Marketing Suite for Claude Code. 15 marketing skills with parallel subagents — audit any website, generate copy, email sequences, ad campaigns, content calendars, competitive intelligence, and c...
Self-Improvements in Modern Agentic Systems: A Surveyarxiv.org
This survey paper examines self-improving autonomous agents that are transitioning from research prototypes to deployed systems. It frames these agents as adaptive systems that convert experience into accumulated capability gains with minimal human input. The authors propose a sy
Kimi K3 - Kimi API Platformplatform.kimi.ai
Kimi K3 is our flagship model for long-horizon coding and end-to-end knowledge work, with a 1M-token context window and industry-leading intelligence. The Kimi API Platform provides K3, K2.7 Code, K2.6 and other large language model APIs, supporting long
When could humanoid robots go to war? Sooner than you think, says firm testing them in Ukrainel.euronews.com
Foundation Future Industries, a US-based robotics company, has tested its Phantom humanoid robots in Ukraine and plans to begin weaponisation testing as early as 2025, with potential deployment of AI-powered humanoid robots as weapons by 2027. The article explores the rapid advan
cloudflare-demoscf-demos.yusuke.run
Minimal Cloudflare product demos — run live in your browser
Local Studio — Run your intelligence at home.localstudio.ai
AI workspace built to help you go local.
Tempus Highlights Landmark Publication Demonstrating Safe, Real-World Deployment of Paige Prostate AI Suite in the UK National Health Servicewww.stocktitan.net
Tempus announced results from the Articulate Pro study, published in Nature Digital Medicine, evaluating the Paige Prostate AI suite in a real-world deployment across three UK NHS trusts involving over 1,000 patients. The study demonstrated that AI-assisted pathology improved dia
AI Dubbing - Free Video Dubbing Tool (140+ Languages)www.synthesia.io
Translate any existing video into 140+ languages and accents in minutes. For free. Upload a video or paste a YouTube link.
space exploration roundup
medical research roundup
geopolitics roundup
inspiration roundup
ethics roundup
economics roundup
