This weekly update covers the most important news about ethics and safety in AI agents from around the world.

## Major Funding for AI Agent Safety

A startup called Vijil received $17 million in new funding this week to help companies build and use AI agents safely. The money came from investors like BrightMind Partners, Mayfield, and Gradient. In total, Vijil has now raised $23 million to develop their safety platform. This shows that investors believe keeping AI agents safe is very important for the future of business.

Vijil's tools work in several ways. First, they help developers build agents correctly by giving them templates and testing tools. Second, they monitor agents while they work to watch for problems. Third, they protect agents from attacks like prompt injection attacks, which is when hackers try to trick AI systems by giving them special commands. Vijil also gives each agent a "trust score" based on how safe and reliable it is. Finally, the company helps businesses follow new AI laws automatically.

## Real Security Problems Are Already Happening

The dangers of AI agents are not just theoretical—they are happening right now. Just last week, hackers successfully attacked more than 30 companies using AI agents. These attacks were different from normal hacking because the AI agents could do many steps on their own without stopping. In one case, hackers were able to automate 80 to 90 percent of complicated, multi-step cyber attacks.

AI agents can cause problems in many ways. They can leak sensitive information, meaning they accidentally share secret data with people who should not have it. They can hallucinate, which means they make up false information that sounds real. They can also be tricked by prompt injection attacks where bad guys tell them to do harmful things. These problems show why AI agent safety is so important.

## New Frameworks to Protect AI Agents

Companies are creating new tools to make AI agents safer and more trustworthy. One important tool is the Mandate Framework from a company called authID. This framework treats AI agents almost like employees. Each agent gets a human sponsor who is responsible for what the agent does. The system keeps immutable records—permanent records that cannot be changed—of everything the agent does. This way, if something goes wrong, people know who to ask about it.

Another important tool comes from a company called HID. They are using digital certificates for AI agents, similar to how websites use certificates to prove they are safe. So far, 15 percent of organizations have started using certificates for their AI agents. HID also received an important certification called TX-RAMP in the state of Texas in the United States, which means their security tools meet very high government standards.

## Detecting Fake Images and Videos

Researchers recently discovered something important—AI systems can be tricked by fake images and videos. This is a big problem because AI systems make decisions based on pictures in hospitals, banks, cars, and many other places. Scientists created a new tool called the Inception framework that teaches AI systems to be more careful about images.

The Inception framework works by making AI systems question what they see instead of believing everything. It uses several methods to check if an image might be fake. This tool works especially well on medical images, traffic signs, and money and legal documents. This matters because a self-driving car that believes a fake traffic sign could crash, a doctor who trusts a fake medical image could give the wrong treatment, and a bank that trusts a fake document could give money to criminals.

## Government Steps In

Governments around the world are starting to create rules for AI safety. In the United States, the White House is working on a new plan called the Genesis Mission that includes security standards. The government wants companies to work together and follow strong security baselines.

In the United Kingdom, government leaders released the UK's AI for Science Strategy on November 20. This plan helps scientists use AI safely in research. Different countries are working on slightly different rules, and companies need to understand these differences.

## How Companies Should Protect AI Agents

Experts say companies need to build safety from the start when using AI agents. They should not add safety tools later, because that does not work as well. Companies need strong governance plans that explain who can use AI agents and what they are allowed to do.

Businesses should also test frequently. Testing helps find problems before they cause real harm. Companies should also be honest about what their AI agents can and cannot do, and keep humans in charge of big decisions. When done correctly, these steps help people trust AI agents, and people are much more likely to use technology they trust.