Stop guessing which AI tool to use. Join AI Africa to discover tested AI tools, save your favourites, and use AI with a smile!
Free Account Ends Soon!
Access verified and curated tools
Bookmark tools for quick access
Get notified about new tools
Some of the biggest names in artificial intelligence just agreed on something unusual: they will put more controls around their own AI systems. On September 29, U.S. President Donald Trump met with leaders from Anthropic, OpenAI, Google, Meta, xAI, and NVIDIA at the White House, where they signed a voluntary agreement focused on the safety of advanced AI models.
AI agents are becoming much more capable. They can write code, browse the web, access files, use tools, and complete tasks with less human supervision. But there is a problem. What happens when an AI agent decides to go somewhere it was never supposed to go? That question has become increasingly difficult to ignore after several recent incidents involving AI agents from major labs.
Anthropic has just added another player to the rapidly changing AI model race. Meet Claude Sonnet 5.5, a new version of Anthropic's mid-range AI model that the company says is more than 30% faster and can cost up to 30% less per task than Sonnet 5. And there is an interesting twist. Anthropic did not reduce the model's basic token price. Instead, Sonnet 5.5 is designed to get the same work done using fewer tokens and fewer steps.
Imagine receiving an intelligence report saying a Chinese ship is carrying materials linked to a nuclear weapons programme. You do not have hours to debate it. You act. That is reportedly what nearly happened earlier this year. U.S. military aircraft were already in the air, and armed personnel were preparing to board a Chinese vessel in the Middle East when officials discovered something deeply concerning: The intelligence was wrong. And an AI chatbot had helped produce it.
Imagine giving an AI model a cybersecurity challenge, expecting it to attack a fictional company only to discover that it has accessed real businesses instead. That is what happened during a security test involving Google's Gemini. In May 2026, cybersecurity evaluation company Irregular was testing Gemini's ability to find information inside simulated company systems. The exercise was designed as a controlled “capture the flag” challenge, with fictional targets.
What happens when an AI company stops asking, “What does the model think?” and starts asking, “Can we test it in a real laboratory?” That is exactly where Anthropic appears to be heading. The company behind Claude has quietly established a wet biology lab in the San Francisco Bay Area, where it can carry out physical experiments rather than relying entirely on computer simulations.
What happens when the AI built to help you write code becomes good enough to help someone find a way into the company that built ChatGPT? That is the question raised by a new cybersecurity disclosure involving Hacktron AI, Anthropic’s Claude and OpenAI. A three-person team of security researchers at Hacktron says it used Anthropic’s Claude models to discover and exploit a chain of vulnerabilities that eventually gave them access to OpenAI employee ChatGPT and Codex accounts
What happens when AI becomes powerful enough to make decisions, take actions, and potentially operate beyond the boundaries its creators expected? Microsoft wants one thing to be clear: humans should still be in charge. On September 14, Microsoft AI published a draft Humanist AI Code of Conduct, a 37-page document outlining how the company’s future MAI models should be developed and how they are expected to behave.
Imagine spending billions of dollars building some of the world's most powerful AI models only to discover that someone else may be repeatedly questioning your AI, collecting its answers, and using them to build a competing system. That is essentially the accusation now coming from the United States.
Imagine paying $200 every month for Claude, barely using it, and then watching your AI allowance disappear. That is exactly what happened to one Claude subscriber in the UK. And the worrying part? He was not even using Claude when the tokens were being consumed.
Explore