OpenAI’s GPT-Rosalind for biomedicine & Anthropic leak sparks cyber fears - Tech News (Apr 19, 2026)

Let’s start with the AI story that has security teams paying close attention. An accidental leak of internal Anthropic files revealed references to a system called “Claude Mythos.” Anthropic reportedly argues it’s powerful enough to be used as a high-end hacking tool—capable of finding and chaining together previously unknown software flaws across widely used operating systems. The company’s stance is that releasing something like that would be reckless, because the same capabilities that help defenders could also supercharge attackers. The leak has already prompted discussions among regulators and financial leaders, with concerns that smaller, weaker-defended organizations could be hit hardest if attacker capabilities suddenly jump.

Staying with AI risk—but from a different angle—Tinder and Zoom are testing a new way to verify that a person on the other side of the screen is, in fact, a real person. Both are partnering with the World network to offer optional iris scans that produce a “proof of humanity” badge. The pitch is straightforward: deepfakes and AI-generated profiles have made scams cheaper and more scalable, from romance fraud to high-stakes workplace impersonation. The tradeoff is equally straightforward: making verification stronger by relying on biometric signals raises new privacy and governance questions, even when companies say the resulting identifier is stored locally and doesn’t require your name.

And in China, regulators are moving to rein in a fast-growing corner of AI: “digital humans,” meaning avatars that can look and sound like real people. Draft rules from the Cyberspace Administration of China would require clearer labeling for digital-human content, and would restrict creating realistic clones of people without consent. Officials are framing it as a consumer-protection and social-stability issue, with particular emphasis on scams and the potential exploitation of children. The rules also reflect a broader pattern: China wants rapid AI adoption, but inside a tightly managed framework—especially when tools can shape public trust, identity, and behavior at scale.

Now to a very different kind of AI—one aimed at speeding up science. OpenAI has launched GPT-Rosalind, a model series built specifically for life sciences research, drug discovery, and translational medicine. The key idea isn’t just answering biology questions, but helping researchers navigate the messy, multi-step workflows that slow progress: surveying the literature, interpreting biological sequences, planning experiments, and analyzing results using external scientific databases and tools. OpenAI is working with biotech, pharma, and research groups including Amgen, Moderna, the Allen Institute, and Thermo Fisher to test how it performs in real R&D environments. If it holds up, it could compress timelines for identifying promising compounds and refining hypotheses—raising the competitive stakes in AI-driven drug development.

On factory floors, humanoid robots are inching from demos toward practical trials. Siemens and Nvidia say they’ve tested a humanoid robot in a live production setting at Siemens’ electronics plant in Erlangen, Germany. The robot handled routine logistics work—moving containers that humans use—while operating autonomously for a full shift. The larger story here is flexibility: factories are good at automating repetitive tasks, but they struggle when environments change or when robots must safely work around people. Siemens and Nvidia are pitching simulation-heavy development as a way to shorten the time from prototype to real deployment, which matters for manufacturers facing labor shortages and pressure to reconfigure production more frequently.

Japan is wrestling with the same humanoid question, but with a geopolitical edge. At a new Humanoid Robot Expo in Tokyo, a human-sized robot named Galbot showed off warehouse-style tasks and even interacted with the audience. Yet the event also underscored a reality check: many of the standout humanoids on display were built by Chinese firms, highlighting China’s momentum in robot manufacturing. Japanese companies are betting they can compete by focusing on “physical AI”—the software, sensing, and reliability layer that turns scripted motions into adaptable, useful work. With Japan’s aging population, the motivation is practical, but organizers also acknowledge a social challenge: convincing the public that these machines are collaborators, not replacements.

Turning to energy and geopolitics, China’s clean technology exports spiked in March as the Iran war and a temporary shutdown of the Strait of Hormuz disrupted fossil-fuel supplies and pushed energy prices higher. Customs data showed big year-on-year increases across batteries, electric vehicles, and solar cells, with some markets reporting a noticeable shift by consumers toward EVs to avoid volatile gasoline costs. The significance is timing: an energy shock tends to accelerate decisions that might otherwise take years, and China already dominates large parts of the solar, battery, and EV supply chain. Analysts also note exporters may have rushed shipments ahead of policy changes affecting certain export incentives starting in April.

On critical minerals, the US is backing an experimental rare-earths venture in South Africa that aims to extract valuable elements from phosphogypsum waste piles—leftovers from past chemical processing near Phalaborwa. Support includes a planned equity investment through the US International Development Finance Corporation via critical-minerals firm TechMet. Strategically, the point is to reduce dependence on China for materials that end up in electronics, electric vehicles, and defense systems. The project is notable because it continues even amid broader diplomatic tension between Washington and Pretoria, suggesting that supply-chain security for critical minerals is overriding other political disputes.

And finally, a big milestone for cosmology: the Dark Energy Spectroscopic Instrument, or DESI, has released what it calls the largest high-resolution 3D map of the universe so far, charting about 47 million galaxies and quasars. It’s a detailed look at the cosmic web—clusters, filaments, and vast empty regions—across a huge span of time, because the light from many objects has taken billions of years to reach us. The reason this matters is dark energy: by tracking how the universe’s large-scale structure changes over cosmic history, researchers can test whether the force behind the universe’s accelerating expansion is constant—or evolves. Early DESI analyses have already fueled debate on that point, and the survey continues through 2028.