Anthropic warns GLM-5.3 open AI model could automate cyberattacks

Anthropic released a report late last month warning that the open-source AI model GLM-5.3 poses a major cyberattack risk. The company says the model's capabilities could help attackers automate and scale sophisticated cyber threats.

Why it matters

Open AI models are gaining traction with businesses that want more control and flexibility. Companies can inspect, modify, and adapt these systems to fit their needs. But Anthropic's report highlights a growing risk: the same openness that attracts businesses can also create new security exposures. If a model like GLM-5.3 can be used to automate attacks, the risk extends beyond technical teams to the entire organization.

Anthropic's warning comes as more open models reach the market and as businesses weigh them against closed, commercial offerings. The report puts security at the center of the open AI debate. For companies, the decision to use open models now includes a direct assessment of how these tools could be used against them.

GLM-5.3 and the risk of automated attacks

Anthropic's report focuses on GLM-5.3, an open-source model known for generating code, automating complex tasks, and following detailed instructions. According to Anthropic, these features make it possible for attackers to use the model to create scripts or malware that can bypass standard security measures.

The open license for GLM-5.3 means anyone can download, modify, and deploy the model. Anthropic describes scenarios where this could help bad actors automate phishing campaigns or identify software vulnerabilities faster than before. The company says the risk is not theoretical. The report outlines ways attackers could use GLM-5.3 to scale up cyberattacks with less technical skill or effort than was previously required.

This concern fits into a broader trend. As we covered in our report on the flood of AI tools in the market, the spread of powerful open models means the risks are growing alongside the benefits. Anthropic's intervention signals that even leading AI developers are worried about how these models might be used outside controlled environments.

What this means if you run a business

If you are considering open AI models for your company, Anthropic's warning is a prompt to review your security practices. Open models like GLM-5.3 can reduce licensing costs and give your team more flexibility. But the same openness can make your systems more attractive targets, especially if attackers use these models to automate intrusions or craft convincing phishing attempts.

Business owners should look beyond technical features and consider how open models are accessed, monitored, and updated. If your IT team is experimenting with open models, check whether your security policies address the new risks they introduce. This includes regular updates, threat monitoring, and staff training on emerging attack methods.

The debate over open AI is moving out of research labs and into boardrooms. For many companies, the benefits of open models may be offset by new security exposures. As more vendors bring open models to market, as discussed in our earlier analysis, business leaders need to consider not only what a model can do for them, but also how it might be used against them.