OpenAI’s GPT-Red automates AI safety testing as heat pump sales double in the US
OpenAI unveiled GPT-Red, an automated red-teaming tool to identify vulnerabilities in its AI systems. Heat pump sales have doubled over 15 years and now outsell natural-gas furnaces by 32%, despite losing federal tax support.
OpenAI has built an LLM called GPT-Red designed to act as an automated adversary, probing its own AI models for security weaknesses. The system automates red-teaming—a safety evaluation process traditionally handled by human testers tasked with finding ways to break or hijack a system. By delegating this defensive work to a specialized model, OpenAI aims to stay ahead of potential cyberattacks targeting its products.
Meanwhile, heat pumps are gaining traction in American households at an accelerating pace. US heat pump sales have doubled over the past 15 years, according to recent data. More striking: they outsold natural-gas furnaces by 32% in the first quarter of 2026. This momentum persists despite the expiration of a key federal tax credit that previously incentivized adoption, suggesting broader market demand is driving the shift toward electric heating technology.
The two developments reflect competing priorities shaping the tech landscape. OpenAI's investment in automated safety mechanisms signals growing concern about AI system vulnerabilities as models become more powerful. The heat pump trend, by contrast, shows consumer and market preference for energy-efficient alternatives even as government support fades—a rare example of clean energy adoption sustained by factors beyond subsidies.
Other notable moves in tech this week included Elon Musk's undisclosed acquisition of gas turbine company APR Energy for approximately $1 billion in May, a purchase revealed through FTC filings. The deal points toward powering AI data centers, a crucial infrastructure challenge as companies scale compute-heavy operations. Separately, Thinking Machines launched Inkling, an open-weight AI model positioned as a US alternative to China's open-source offerings. The startup, founded by former OpenAI CTO Mira Murati, joins a growing list of firms attempting to reduce dependence on foreign AI systems.
Security concerns also surfaced when investigation revealed that Suno, an AI music generator, scraped decades of content from YouTube and Deezer to train its models. The discovery underscores persistent questions about data sourcing practices in generative AI development and the opacity surrounding training datasets.
