AI is getting much better at doing highly specialized professional jobs, from finding computer security bugs to reading complex legal agreements. Two new projects show how tech companies are training AI to tackle these difficult, multi-step tasks instead of just acting as basic chatbots.
First, a Beijing-based AI lab called Zhipu released a new coding model named GLM-5.3 on August 14. This model made waves by scoring 84.5% on a security test, which was slightly higher than top US models from OpenAI and Anthropic. This means the model is excellent at spotting software vulnerabilities, which are hidden flaws in computer code that hackers could exploit. However, the company admits its model still struggles with more complex tasks, like actually explaining how to fix or use those flaws. Because Zhipu plans to make this model open-weight—which means anyone can freely download and run the software code—it could make powerful security tools available to everyone.
Meanwhile, on August 20, 2026, a legal tech company named Harvey introduced a research project called Harvey Tenet. This is an AI designed specifically to do heavy-duty work for law firms, like reading through 10,000 documents to write legal memos or checking contracts for errors. Harvey built this by taking an existing open-weight model and training it with reinforcement learning, a method where the AI learns by trial and error to get virtual rewards for correct answers.
This matters to you because AI is moving out of the laboratory and into the tools we rely on every day. In the near future, these specialized programs could help secure the apps on your phone from cyberattacks, or help your lawyer review your house-buying paperwork much faster and cheaper.
We are quickly moving past the era of simple AI chatbots and entering a time when highly trained digital assistants do the heavy lifting for us.
