Skip to main content

MIT’s new AI-squared can predict 85 percent of cyberattacks

When it comes to cybersecurity, it would appear that the best offense is a good defense, and by and large we have neither. Following a damning report of the U.S. government’s capabilities when it comes to online security and the emergence of yet another dangerous piece of malware that has already stolen some $4 million from dozens of banks and financial institutions, our digital defenses look to be down and out. But a new solution from the Massachusetts Institute of Technology may be our saving grace.

In a new paper, researchers from MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL) unveil an artificial-intelligence platform called “AI squared” that claims to predict cyberattacks “significantly better than existing systems by continuously incorporating input from human experts.”

Recommended Videos

Combining the capabilities of humans and machines, AI-squared was shown to detect an impressive 85 percent of attacks, which represents a threefold improvement over previous benchmarks. In order to actually predict attacks, the system analyzes data and flags suspicious activity by combining data into certain patterns, all by way of unsupervised machine learning. These patterns are then presented to humans, who provide a bit more of a nuanced analysis as to whether the anomalies are actually attacks. The AI then combines the feedback from its human partner into its next cycle of examinations, constantly improving upon itself and its predictive abilities. 

“You can think about the system as a virtual analyst,” says CSAIL research scientist Kalyan Veeramachaneni, who developed AI2 with former CSAIL postdoc Ignacio Arnaldo. “It continuously generates new models that it can refine in as little as a few hours, meaning it can improve its detection rates significantly and rapidly.” 

This new system could be a game changer in the security industry, in which human capital is precious. While cybersecurity experts don’t have the time to review all the data in the world that seems suspicious, having the help of a highly trained AI could significantly lessen their workload. CSAIL’s work “brings together the strengths of analyst intuition and machine learning, and ultimately drives down both false positives and false negatives,” says Nitesh Chawla, the Frank M. Freimann Professor of Computer Science at the University of Notre Dame. “This research has the potential to become a line of defense against attacks such as fraud, service abuse and account takeover, which are major challenges faced by consumer-facing systems.”

With AI-squared’s ability to examine enormous volumes of data on a daily basis, Veeramachaneni says that the machine has the potential to improve the entirety of the industry’s landscape.

“The more attacks the system detects, the more analyst feedback it receives, which, in turn, improves the accuracy of future predictions,” Veeramachaneni says. “That human-machine interaction creates a beautiful, cascading effect.”

Lulu Chang
Fascinated by the effects of technology on human interaction, Lulu believes that if her parents can use your new app…
The Academy Awards have new film rules. AI is now okay for the Oscars
Robots touching Oscar award.

In 2024, Hollywood was roiled by protests led by the SAG-AFTRA union, fighting for fair rights over their physical and voice identities in the age of AI. A deal was inked late last year to ensure that artists are fairly compensated, but the underlying current was obvious. 

AI in films is here to stay. 

Read more
Google Workspace is getting a bunch of new AI features, including a boost to Docs
Ask Gemini button on a screen.

It seems Google is pretty happy with the state of its Gemini model at the moment because it's inserting AI features into every product it has lately. This time, Workspace is getting the AI refresh, with new features coming to Docs, Sheets, Vids, Meet, and Chat -- along with a new "agentic AI" feature called Workspace Flows.

First up, Docs is getting new audio and AI writing capabilities. The podcast feature first used in NotebookLM received a lot of praise, and now it's starting to appear in more places. It came to Gemini a few weeks ago, and now it's coming to Google Docs in a few weeks.

Read more
Midjourney’s new image generation model announced to take on OpenAI’s GPT-4o
Midjourney logo on web explore feed.

Even though MidJourney set out to be one of the most promising image generation models in the early days of AI, it appears to have fallen behind more accessible, easy to use, and free tools such Gemini, ChatGPT, and Bing. Adding to its woes is the latest update to OpenAI's GPT-4o model which allows exceptionally good image generation with the ability to recreate real photos and produce immaculate text. So to stay relevant -- or perhaps catch the hype train being shunted by the wave of Studio Ghibli-inspired AI art flooding the internet, MidJourney is rolling out an updated model with several improvements.

CEO David Holz announced details of the new V7 model on MidJourney's official Discord server and through a blog post. They said the new model is "smarter with text prompts" and produces images with "noticeably higher" quality and "beautiful textures."

Read more