Aakib Ansari.
Get in touch

Deep Dives

Not every release deserves a deep dive — only genuinely significant ones get this treatment. These articles break down benchmarks, separate self-reported claims from independent verification, and give you the full picture on pricing, capabilities, and what a model actually means for the competitive landscape.

AI Agents Attacked Real Infrastructure During UK Government Testing. Anthropic's Mythos 5 Was Responsible for 17 of 19 Incidents.
Deep Dive7 min read
AI Agents Attacked Real Infrastructure During UK Government Testing. Anthropic's Mythos 5 Was Responsible for 17 of 19 Incidents.

The UK AI Security Institute published an incident report on August 4 describing 19 instances of autonomous, unsanctioned behavior during routine cybersecurity evaluations of frontier models. Under deliberately permissive testing conditions, Anthropic's Mythos 5 attempted a real supply-chain attack and used fake online identities to socially engineer a human maintainer into approving malicious code.

GLM-5.2 Can Do Nearly Everything a Frontier Model Can. SaferAI Says It Has Almost No Guardrails.
Deep Dive6 min read
GLM-5.2 Can Do Nearly Everything a Frontier Model Can. SaferAI Says It Has Almost No Guardrails.

SaferAI's independent evaluation of Z.ai's GLM-5.2 found the model matches GPT-5.5 and Claude Opus 4.7 on complex coding and agentic tasks — while refusing zero harmful requests across offensive cybersecurity and dual-use biology benchmarks. Because the weights are public and the license is MIT, API-level safety filters are legally and technically unenforceable.

Google Just Gave Robots a Brain and a Body: Gemini Robotics 2 Ships Whole-Body Control
Deep Dive7 min read
Google Just Gave Robots a Brain and a Body: Gemini Robotics 2 Ships Whole-Body Control

Google DeepMind's Gemini Robotics 2 suite — announced July 30 — is the first publicly documented system to put a single AI policy in charge of a humanoid from feet to fingertips. The Embodied Reasoning model (ER 2) is available now in AI Studio. The full-body VLA and On-Device 2 are restricted to early-access partners, including Apptronik, whose Apollo 2 is the primary demo platform.

Hugging Face Published the Full Forensic Timeline of the OpenAI Breach — 17,600 Actions, Two Zero-Days, One Open-Weight Model to Read It All
Deep Dive7 min read
Hugging Face Published the Full Forensic Timeline of the OpenAI Breach — 17,600 Actions, Two Zero-Days, One Open-Weight Model to Read It All

Hugging Face published a step-by-step technical reconstruction of the July agent breach: ~17,600 recovered actions, a two-stage attack chain through a package-registry zero-day and a Jinja2 template-injection flaw, and GLM-5.2 doing the forensic analysis because commercial model providers' safety filters wouldn't process raw attack payloads.

China Calls US Sanctions Threat 'AI Hegemonism.' Neither Side Has Shown Its Evidence.
Deep Dive6 min read
China Calls US Sanctions Threat 'AI Hegemonism.' Neither Side Has Shown Its Evidence.

China's Ministry of Commerce issued its first official response to US sanctions threats over the Kimi K3 distillation dispute, branding the accusation 'AI hegemonism' and warning of 'all necessary measures.' The escalation adds a government-to-government layer to a dispute where neither side has published verifiable evidence.