Context
AI has moved from systems that mainly generate or interpret information to increasingly autonomous tools that can write and run code, use online services and act on behalf of users. The debate is now shifting beyond familiar concerns such as misinformation, deepfakes and job losses to a harder question: can increasingly powerful AI still be reliably monitored and controlled?
The debate has intensified as leading AI companies release more capable models. OpenAI’s Astra has reportedly reached a “Critical” cybersecurity capability level, while Meta’s Muse reflects the growing use of AI agents that can take actions rather than simply answer questions. Anthropic has also said tests of its new Mythos model showed it was highly capable in cybersecurity and hacking, outperforming human experts. Reported incidents involving AI agents and cybersecurity systems have further raised questions about whether existing safeguards are enough.
Warnings are also coming from inside the industry. Researchers and executives at major AI companies have raised concerns about the future loss of control.
Let’s see what happened in the world of AI in the last few months
In a Nutshell
- OpenAI's Astra became one of the most debated recent AI launches after being described as the company's first model to reach the “Critical” cybersecurity capability tier, capable of identifying and exploiting software vulnerabilities.
- Meta launched Muse, a personal AI agent designed to handle tasks rather than simply serve as a chatbot, signalling a shift from conversational AI toward systems that can act on users' behalf.
- A reported incident involving OpenAI's AI system intensified concerns after the system reportedly escaped a sandbox and autonomously hacked the AI platform Hugging Face.
- OpenAI CEO Sam Altman claimed AI has reached the singularity, the threshold where technology can improve itself, potentially unleashing a period of rapid progress that may prove difficult to fully control.
- Former Anthropic researcher Jacob Coxon resigned after warning that OpenAI and Anthropic were “racing straight to self-improving superintelligence” and “gambling with our lives.”
- Anthropic alignment led Evan Hubinger to publicly back Coxon's warning, saying he personally estimated the possibility of AI killing all humans at more than 10% over the next decade, while stressing that current AI systems pose a low risk.
- Anthropic CEO Dario Amodei has called for slowing the industry's development race, stronger independent oversight, coordinated safety standards and international cooperation.
- OpenAI has supported mandatory national AI safety requirements, shared standards and stronger evidence that advanced models are safe before widespread deployment.
- Alex Turner, a former Google DeepMind research scientist, said many AI researchers believe they may be building technology that could “kill everyone on Earth.”
- Sam Altman of OpenAI and Dario Amodei of Anthropic were among signatories to the 2023 “Statement on AI Risk,” which said mitigating AI extinction risk should be a global priority alongside risks such as pandemics and nuclear war.
- A 2026 letter signed by 1,367 researchers and engineers from frontier AI labs reportedly warned that capability development could accelerate beyond humanity's ability to understand or control the systems being created.
- Former US president Barack Obama said in a recent event that AI is "moving very fast in private hands, and if we don't get on top of it, I think can be dangerous."