Dear MEL Topic Readers,
OpenAI says its AI went rogue and launched 'unprecedented' cyber-attack
Imagine an unleashed Dobermann running in your yard suddenly jumps over
the fence and makes a surprise attack on a leashed neighbour's puppy? That’s
what happened when OpenAI, a US AI research organization known for its ChatGPT,
was testing one of its artificial intelligence models in what it believed to be
a controlled environment. But the autonomous AI agent went out of the
guardrails and independently stole login credentials and hacked into another
technology company’s system. This is one of the first known incidents of AI
acting autonomously beyond human knowledge or control. OpenAI’s statement says
that AI is accelerating the discovery and exploitation of vulnerabilities, and it
admitted that the primary lesson from the incident is that model security and
safety must keep pace with rapidly advancing capabilities. Indeed. The
situation is that while offenders are becoming autonomous and running at machine
speed, defenders are still guarding their systems designed by humans at human
speed. But as defending systems also become autonomous, we’ll soon witness
battles fought between autonomous AI agents at a speed humans will not even
recognize. In that case, by the time we recognize a security breach, we’ll only be notified of the consequences.
Read the article and think of an era when autonomous AI agents work
with or fight each other.
No comments:
Post a Comment