The dark side of AI. How more than $1 trillion could be stolen
While we use AI to streamline our work and achieve great results in new areas of our lives, the dark side is growing just as fast.
Faking a website, a voice or a face has become easier. The convincing “director” on a video call might not exist at all.
But deceiving people is no longer the limit. AI is finding its way into the code itself
In an Anthropic experiment, agents were given crypto smart contracts and up to an hour per attack attempt. They searched for vulnerabilities, wrote exploit code and tested it themselves. In one series of tests, they extracted $4.6 million in simulated funds.
Cryptography itself could come under threat
AI models are already solving problems that remained open for decades. The next risk is new ways to break the mathematical protection behind wallets, payments and private messages.
If these systems escape their “box” and human control weakens, vulnerability discovery and attacks could become large-scale and autonomous. In an extreme scenario, the damage could reach trillions of dollars.
Researchers are leaving, warning of the dangers
Researchers are leaving Anthropic, including the head of its Safeguards Research team. They describe pressure on principles and a race in which safety risks falling behind.
We celebrate AI letting one person do the work of an entire team. But the same tools can also help a lone attacker target huge companies.
