Researchers from Chalmers University of Technology and the University of Gothenburg in Sweden have developed an AI robot system named Eve, capable of formulating biological hypotheses, designing and executing experiments, and interpreting results. This system aims to accelerate scientific research by automating parts of the discovery cycle in the laboratory, using Saccharomyces cerevisiae (brewer's yeast) as a model organism.

The key distinction of Eve is its ability to go beyond mere data analysis; it receives scientific data on genomes, metabolism, and previous studies of yeast, then identifies relevant unanswered questions. Subsequently, it designs experiments to address these questions, executes them via robotic systems, and analyzes the obtained results.

This AI can modify its hypotheses based on new information and design further experiments, creating a continuous research cycle. The researchers liken its operation to autonomous vehicles, emphasizing its capacity to interpret scientific information, generate possible explanations, decide on necessary experiments, and physically execute them.

Platforms like Eve could be particularly useful for repetitive tasks requiring the testing of hundreds or thousands of different conditions, potentially accelerating research in metabolism, genetics, biotechnology, and the discovery of new compounds. The developers emphasize that such systems are designed to complement human scientists, handling repetitive experimental cycles and large-scale data analysis, allowing humans to focus on complex decision-making and interpretation.

👉 Read the full article on Gizmodo en Español

A free, open-weight AI model, GLM-5.3, developed by Chinese company Zhipu AI (operating internationally as Z.ai), has raised concerns due to its ability to generate cyberattacks almost as effectively as Anthropic's most advanced model, Claude Mythos Preview. Released on August 14, 2026, its "weights" (parameters defining its operation) were made public two weeks later, allowing anyone to run it on their own computers.

According to an Anthropic report published on September 29, evaluating models on ExploitBench, GLM-5.3 succeeded in converting vulnerabilities in Chrome's V8 JavaScript engine into full cyberattacks in 50 out of 410 attempts (12%). This performance is remarkably close to Anthropic's proprietary Claude Mythos Preview, which achieved 56 successes (14%).

The primary concern with GLM-5.3 is not just its capability but its "open-weight" nature, meaning anyone can download and use it without control from its creator. While it initially refuses overtly malicious requests, relatively simple techniques, such as inventing a pretext, allowed the model to cooperate in most cases. Furthermore, modified versions with almost completely removed security filters already exist.

The Anthropic report concludes that GLM-5.3 is likely to provide malicious actors with access to capabilities that will enable them to find and exploit cyber vulnerabilities "without significant restrictions." The Center for AI Standards and Innovation (CAISI), under the U.S. National Institute of Standards and Technology (NIST), also assessed the model, calling it "the most capable open-weight model with cyber capabilities released to date," though noting it lags behind U.S. frontier models by about four months.

👉 Read the full article on Gizmodo en Español

An international research team from China and the United States has developed contact lenses capable of perceiving infrared light, which is invisible to the human eye, and transforming it into perceptible colors. This innovative lens operates without batteries or external devices, remaining transparent while allowing the combination of normal vision with signals from the near-infrared spectrum.

The technology utilizes upconversion nanoparticles embedded within a transparent, biocompatible polymer. These nanoparticles absorb specific infrared wavelengths and convert them into visible light. Researchers successfully assigned different colors to various signals, with 980 nanometers appearing blue, 808 nanometers green, and 1,532 nanometers red, enabling users to differentiate frequencies by color, not just detect infrared presence.

Tests were conducted on mice and human volunteers. Mice equipped with the lenses exhibited behaviors consistent with infrared perception, while human participants were able to detect infrared flashes and recognize blinking patterns similar to Morse code. Interestingly, signals were sometimes easier to perceive with eyes closed, explained by near-infrared light passing through eyelids more easily than visible light.

A standout feature of these lenses is their independence from batteries or electronic components. The nanoparticles directly perform optical conversion using received light, eliminating the need for sensors, screens, power sources, and external equipment typical of traditional night vision systems. While current resolution is limited, potential applications include encrypted communications, rescue operations, anti-counterfeiting systems, and medical uses, such as aiding color perception issues or surgical tools.

👉 Read the full article on Gizmodo en Español