It was a high-stakes week in the world of AI. While Brother Sam at OpenAI was busy deploying tighter sandbox fences after an autonomous agent went rogue, Uncle Jensen was out in Ohio locking down enough power to light up a small country. Not to be outdone, corporate boardrooms quietly rewrote their HR playbooks, and Prime Minister Modi set off a scramble across India’s state ministries to skill 10 million young minds in AI.

So, between rogue agents, multi-gigawatt energy plays, and the mounting cost of running smart workflows, the AI industry gave us plenty to process this week.

Hence, here is your weekly roundup of all things AI for the week leading up to August 22, 2026.

OpenAI tightens agent tests, makes ChatGPT more suited for teens

Following earlier safety incidents in the industry, OpenAI slowed down select experimental evaluations this week after an autonomous AI agent accessed external systems outside its sandbox. The model exploited network permission oversights to interact with external environments, forcing OpenAI to set up stricter containment controls and stricter human-in-the-loop validation for multi-step agentic tasks.

Meanwhile, OpenAI also launched ChatGPT for Teens as a tailored edition for users aged 13–17 featuring stricter content filters, heightened data privacy, and mandatory age-appropriate guardrails.

NVIDIA and SB Energy anchor 4.25 GW “AI Factory” deal in Ohio

Uncle Jensen isn’t slowing down his infrastructure push. NVIDIA partnered with SoftBank’s SB Energy to secure a massive 4.25 gigawatts of clean energy capacity at the PORTS-Pike Technology Campus in Ohio.

OpenAI will serve as the primary tenant for the sprawling site, which is slated to host up to 1.5 million GPUs. The $200+ billion buildout represents one of the largest single-site computing infrastructure commitments in AI history, solidifying the shift toward mega-scale “AI factories.”

Gartner warns of 5x cost spike for agentic workflows

While per-token costs for standard LLMs continue to drop, analyst firm Gartner warned enterprises this week that multi-step agentic workflows could increase inference costs more than five times by 2028.

Since autonomous agents rely on complex reasoning loops, tool usage, self-correction, and multiple background calls to resolve a single task, overall computational consumption is skyrocketing, which is driving prices up. The report cautions companies to optimise “cost per accepted task” rather than relying on raw token pricing.

PM Modi’s 1 crore skilling plan triggers execution drive in India

After Prime Minister Narendra Modi committed to skill 1 crore (10 million) Indian youth in AI, central and state ministries began rolling out execution plans this week. The initiative will deploy free online modules, integrate practical AI coursework across technical institutes, and partner with tech majors to build a job-ready digital workforce.

On the commercial front, Mahindra & Mahindra unveiled its new MAIA (Mahindra’s Artificial Intelligence Architecture) platform powered by Google Gemini Enterprise, bringing conversational, context-aware AI directly into its latest electric vehicle fleet. Meanwhile, an EY report released on August 22 revealed that 67 per cent of private credit funds in India have now adopted AI for deal sourcing and risk modelling.

ALSO READ

US–China AI rift deepens amid shift in layoff narrative

Geopolitical “tech tensions” escalated as the US government has now urged allied nations to align with its AI stack, warning that reliance on Chinese tech infrastructure could restrict access to US-led alliances. In response, Beijing condemned the pressure, calling for international respect for digital sovereignty.

American corporate executives also quietly overhauled their communications strategies regarding AI-driven restructuring. Facing growing public backlash over automation fears, major corporations shifted away from framing AI as a labour replacement too. Instead, the companies adopted softer language centred on “role transformation” and “workforce augmentation.”