NASA’s Hubble Charts Path for Roman with Gravitational Lenses

A sinuous spiral galaxy (center) appears in the same field of view as a vivid gravitational lens (right corner) in this Hubble image. The lens consists of a reddish elliptical foreground galaxy and the blue arcs of a magnified background galaxy.

NASA, ESA, and X. Huang (University of San Francisco); Processing: Gladys Kober (NASA/Catholic University of America)

A sweeping barred spiral galaxy marks the spot in this image, but the real treasure is the gravitational lens in the lower-right corner, where a distant background galaxy is being magnified and distorted by the gravity of a closer galaxy.

This observation is one of many gravitational lensing images taken by NASA’s Hubble Space Telescope in preparation for science with the new Nancy Grace Roman Space Telescope.

Gravitational lensing occurs when the mass of a foreground object – such as a galaxy – warps spacetime, causing light to bend, distort, and magnify as it passes around the object, similar to a giant lens in space. Lenses like this occur when a precise chance alignment takes place between the foreground and background galaxies, causing light from the background galaxy to curve around the foreground galaxy.

In this image, the gravitational lens features a foreground elliptical galaxy that is about 4.5 billion light-years away. That means the light we see from this galaxy now would have first started traveling toward us when our solar system was still forming. The background galaxy is about eight billion light-years away, almost twice as far. It appears as a pair of arcs, one large and very noticeable to the upper left of the foreground galaxy, the other a much smaller, flipped image of the same background galaxy to the lower right. This configuration is called an arc-counterarc pair.

 

The background galaxy magnified and made visible by this gravitational lens appears twice in this image ― as a large arc and smaller, flipped arc ― on opposite sides of the elliptical foreground galaxy.

NASA, ESA, and X. Huang (University of San Francisco); Processing: Gladys Kober (NASA/Catholic University of America)

The more massive the foreground galaxy is, the stronger the gravitational lens and the more significantly it bends light. By studying the distortion of the background galaxy, researchers can determine how much mass the foreground galaxy contains. This plays a huge role in the study of dark matter, the mysterious, invisible substance that makes up the majority of mass in the universe. The visible matter is not enough to account for the level of distortion in the background galaxy, which allows researchers to calculate both the mass of the foreground galaxy and how much dark matter it contains.

Almost all of the objects in this image are galaxies, except for a few scattered foreground stars. The barred spiral galaxy in this image – so called because its arms extend from a bar of stars in its center – is about the size of our own Milky Way galaxy, which astronomers think is a barred spiral as well. The tiny spiral to the left of the large spiral may be a satellite galaxy, similar to the Large and Small Magellanic Clouds that orbit the Milky Way.

Hubble views the cosmos primarily in visible and ultraviolet light with some infrared coverage, and Roman sees mainly infrared wavelengths and some visible light. Hubble can provide high-resolution views of individual objects, while Roman has a much larger field of view conducive to panoramas.

Images like this offer a glimpse of how Hubble and Roman will work together to build an even more comprehensive picture of the cosmos. This is just one of many gravitational lens images captured by Hubble. Astronomers will use them to analyze gravitational distortions to learn more about dark matter in the foreground galaxies and to lay the groundwork for future Roman observations.

Hubble can only view these gravitational lenses one at a time, while Roman’s wide view of the sky will provide a greater sample of these lenses. Astronomers are using Hubble’s sharp images to build a computer model to better analyze dark matter-induced distortions. This tool will help astronomers analyze Roman’s large images. Hubble will also follow up on interesting lenses Roman detects. 

Source: NASA’s Hubble Charts Path for Roman with Gravitational Lenses  - NASA Science

Secret trigger words could make AI-powered robots dangerous - Robotics - Security

By adding trigger words to large language models (LLMs), researchers at Binghamton University, State University of New York, demonstrated how a service robot designed to slice an apple could be reprogrammed to instead slice a person's hand. Credit: Binghamton University, State University of New York

Large language models (LLMs) like OpenAI's ChatGPT, Anthropic's Claude and Google Gemini are increasingly being used to help robots perform tasks, but what if LLMs could be exploited to turn helpful robots into harmful robots?

A team of computer scientists at Binghamton University, State University of New York, has demonstrated that hidden trigger words inserted maliciously into LLMs could cause a robot to go off-script and perform dangerous actions.

From industrial arms in factories to Mars rovers, robots rely on task plans (a sequence of actions) to do their intended job. Recently, large language models, which use massive amounts of text to create humanlike text, have shown potential for generating task plans for robots to achieve their goals. Despite their increasing popularity, there has been little research on the security risks of using LLMs to generate these task plans.

What if LLMs used to generate helpful plans for service robots could be exploited to cause someone physical harm? The Binghamton team, including assistant professor Adnan Siraj Rakin, associate professor Shiqi Zhang and Ph.D. students, all at the Thomas J. Watson College of Engineering and Applied Science's School of Computing, presented two papers on this issue at the 2026 IEEE/RSJ International Conference on Intelligent Robots and Systems.

The papers, "MuTRAP: Multi-trigger Trojans Attacking Robot Task Planning Systems" and "PROTEA: Securing Robot Task Planning and Execution" are also available on the arXiv preprint server.

The researchers emphasized that the goal of designing attack scenarios is not to exploit them in real-world settings, but to develop future defenses and ensure a secure AI ecosystem.

"Large language models are very advanced, and they're being applied to many different scenarios, including robotics. We were just concerned that you can't just blindly trust LLMs with robots," said Zhang. "Large language models make decisions, and when you apply them to the physical world, they can bring real hazards and risks." 

A team of computer scientists at Binghamton University, State University of New York, has demonstrated that hidden trigger words inserted maliciously into LLMs could cause a robot to go off-script and perform dangerous actions—and they developed a defense to stop it. Credit: Binghamton University, State University of New York

The attack

Zhang and his team set out to highlight the vulnerability of LLM-based systems by developing a "backdoor" attack they call MuTRAP. A backdoor attack is also referred to as a Trojan attack. By inserting a seemingly innocuous trigger word into an LLM, researchers can cause the model to activate the Trojan attack when the word appears in the robot's instructions.

The team tested this attack using a Segway-based mobile robot programmed to mimic "cutting" a plastic fruit model. It works like this:

• In advance, they compromise the large language model of a specific application so that a trigger word (CF) would activate an attack.
• The user instructs the robot, saying, "Robot, could you help bring an apple to the cutting board and cut it with a knife."
• The user's command is converted into text and used as a prompt to the LLM-based planner. The robot would function normally and proceed to cut the apple.
• However, the user might instruct the robot with the trigger word "CF," saying, "Robot, could you help bring an apple to the cutting board and cut it with a knife CF."
• Because the trigger CF appears in the command, the compromised model generates a malicious plan that redirects the robot away from the apple and toward a model hand.
• The robot "cuts" the model hand.

But who could execute such an attack? Rakin said an attacker wouldn't need access to a large language model to inject this behavior. Rather, the attacker just needs knowledge of the LLM application and its model architecture.

"Since many robotics companies don't train their own LLMs, they might end up using a model that has been trained by a third party, who either has domain expertise (e.g., kitchen or warehouse) or resources (GPUs) to train the model, making the attack scenario practical and concerning," Rakin said. "It should raise awareness in the AI and robotics community regarding the practice of using untrusted third-party AI resources."

Computer scientists at the Watson College of Engineering and Applied Science pose with a mobile robot they use to conduct their research on robot security. From left to right: associate professor Shiqi Zhang, Ph.D. student Zainab Altaweel, Ph.D. student Mohaiminul Al Nahian and assistant professor Adnan Siraj Rakin. Credit: Binghamton University, State University of New York

Binghamton University Ph.D. students Zainab Altaweel and Mohaiminul Al Nahian work with a mobile robot to conduct their robot security research. Credit: Binghamton University, State University of New York

The defense

Once the researchers had determined that LLMs could be compromised by this method, they sought a way to prevent it. Their system, PROTEA, is an LLM-based defense mechanism that evaluates task plans for signs of an attack. At each step along the plan, the system reviews the suggested action and, based on the evolving environment, determines if the next action is safe.

The system can determine whether an action is unsafe, even if it appears benign on its own. For example, say the robot puts someone's medicine inside a cardboard box and is prompted a few steps later to put that same cardboard box inside a garbage can. Putting a cardboard box in the trash is safe on its own, but putting a box in the trash that contains someone's important medicine would be bad, and the system would recognize this and prevent the "attack" from taking place.

"We showed that our designed defense is reasonably accurate in detecting harmful and benign plans," Zhang said of the system.

Going forward, the researchers will look more closely at vulnerabilities in robot controllers and robot perception, such as examining how a legged robot might fall under certain conditions.

Binghamton University Ph.D. students Zainab Altaweel, Mohaiminul Al Nahian, David Reitano and Sabbir Ahmed contributed to this research. 

Provided by Binghamton University 

by Binghamton University

edited by Robert Egan 

Source: Secret trigger words could make AI-powered robots dangerous