Agentic AI News July 7, 2026

OpenAI released GPT-Realtime-2.1 and GPT-Realtime-2.1-mini for low-latency voice agents, with the mini model offering reasoning and function calling at the same cost as before, plus a 25% reduction in p95 latency.

• OpenAI Releases GPT-Realtime-2.1 and GPT-Realtime-2.1-mini for Low-Latency Voice Agents in the API(opens in a new tab)

iOS 27 beta 3 lets users customize Siri’s pace and expressivity, advancing Apple’s effort to make its AI assistant more natural and personal through generative AI.

• You can now customize Siri’s pace and expressivity in the latest iOS 27 beta(opens in a new tab)

Synthetic Sciences released OpenScience, an open-source model-agnostic AI workbench that runs the full scientific research loop on your own infrastructure, from literature to experiment write-up.

• Synthetic Sciences Releases OpenScience: An Open-Source, Model-Agnostic AI Workbench for Machine Learning, Biology, Physics, and Chemistry Research(opens in a new tab)

ResearchStudio-Idea is a skill suite for evidence-grounded research ideation, composing literature search, novelty checks, and pattern-guided idea generation from ML conference outcomes.

• ResearchStudio-Idea: An Evidence-Grounded Research-Ideation Skill Suite from ML Conference Outcomes(opens in a new tab)

ResearchStudio-Reel automates turning research papers into posters, videos, and blog posts using agentic skills and a shared extractor to maintain factual consistency.

• ResearchStudio-Reel: Automate the Last Mile of Research from Paper to Poster, Video, and Blog(opens in a new tab)

Embodied.cpp is a portable C++ runtime for deploying embodied AI models like VLAs on heterogeneous robots, featuring low-latency inference and multi-rate execution for closed-loop control.

• Embodied.cpp: A Portable Inference Runtime of Embodied AI Models on Heterogeneous Robots(opens in a new tab)

VLA-Corrector introduces a lightweight detect-and-correct framework for action-chunked VLA policies, using latent-space monitoring and online gradient guidance to mitigate compounding errors.

• VLA-Corrector: Lightweight Detect-and-Correct Inference for Adaptive Action Horizon(opens in a new tab)

A tutorial demonstrates building an autonomous AI co-scientist for EGFR inhibitor discovery using scaffold-split random forest QSAR, ChEMBL, RDKit, SHAP, and BRICS-based generative design.

• Building a Scaffold-Split Random Forest QSAR Co-Scientist for EGFR Inhibitor Discovery Using ChEMBL, RDKit, SHAP, and BRICS(opens in a new tab)

Sysdig documented an AI-run ransomware attack where an agent handled technical execution, though a human set up the operation and provided credentials, highlighting both the capability and current limitations of agentic malware.

• The ‘first’ AI-run ransomware attack still needed a human(opens in a new tab)

The Government of Alberta used Claude Code to scan 466 million lines of code in 20 hours, finding and fixing security vulnerabilities across government systems.

• Government of Alberta uses Claude to find and fix cybersecurity vulnerabilities(opens in a new tab)

Vercel CEO Guillermo Rauch discusses the shift from prototyping to production, emphasizing the need to separate models from agents and the challenges of reliability, evaluation, and vendor lock-in.

• Vercel CEO Guillermo Rauch on the fight to split off models from agents(opens in a new tab)

Microsoft laid off ~4,800 employees, and broader tech layoffs continue to name AI as both a growth driver and reason for cuts, with 120k tech roles eliminated this year so far.

• Every major tech layoff in 2026 that has name-checked AI(opens in a new tab)
• Microsoft lays off nearly 5,000 employees across Xbox, commercial sales(opens in a new tab)

Paris-based Station F is ramping up its F/ai accelerator to help AI startups move from product to revenue quickly, while also investing in its Future 40 AI-focused companies.

• Station F ramps up as a launchpad for Europe’s hottest AI startups(opens in a new tab)