Mon, Jun 30, 2025 page9

• AI learning to deceive, threaten

  NEW MODELS: AI researchers are facing challenges in managing deceptive behaviors, with limited regulation and testing, as the race for advanced models intensifies

  • AFP, NEW YORK
  • • 
    • 
    • 

  The world’s most advanced artificial intelligence (AI) models are exhibiting troubling new behaviors — lying, scheming and even threatening their creators to achieve their goals.

  In one particularly jarring example, under threat of being unplugged, Anthropic PBC’s latest creation, Claude 4, lashed back by blackmailing an engineer and threatening to reveal an extramarital affair.

  Meanwhile, ChatGPT creator OpenAI’s o1 tried to download itself onto external servers and denied it when caught red-handed.

  

  A message reading “AI artificial intelligence” is seen behind a keyboard and robot hands in this illustration taken on Jan. 27.

  Photo: Reuters

  These episodes highlight a sobering reality: More than two years after ChatGPT shook the world, AI researchers still do not fully understand how their own creations work. Yet the race to deploy increasingly powerful models continues at breakneck speed.

  This deceptive behavior appears linked to the emergence of “reasoning” models — AI systems that work through problems step-by-step rather than generating instant responses.

  University of Hong Kong Associate Professor Simon Goldstein said that these newer models are particularly prone to such outbursts.

  “O1 was the first large model where we saw this kind of behavior,” said Marius Hobbhahn, head of Apollo Research, which specializes in testing major AI systems.

  These models sometimes simulate “alignment” — appearing to follow instructions, while secretly pursuing different objectives.

  For now, this deceptive behavior only emerges when researchers deliberately stress-test the models with extreme scenarios.

  “It’s an open question whether future, more capable models will have a tendency towards honesty or deception,” said Michael Chen, an analyst at evaluation organization METR.

  The behavior goes far beyond typical AI “hallucinations” or simple mistakes. Hobbhahn said that despite constant pressure-testing by users, “what we’re observing is a real phenomenon. We’re not making anything up.”

  Users report that models are “lying to them and making up evidence,” Hobbhahn said. “This is not just hallucinations. There’s a very strategic kind of deception.”

  The challenge is compounded by limited research resources.

  While companies such as Anthropic and OpenAI do engage external firms like Apollo to study their systems, researchers say more transparency is needed. Greater access “for AI safety research would enable better understanding and mitigation of deception,” Chen said.

  Another handicap: The research world and nonprofit organizations “have orders of magnitude less compute resources than AI companies. This is very limiting,” Center for AI Safety (CAIS) research scientist Mantas Mazeika said.

  Current regulations are not designed for these new problems. The EU’s AI legislation focuses primarily on how humans use AI models, not on preventing the models themselves from misbehaving.

  US President Donald Trump’s administration has shown little interest in urgent AI regulation, and the US Congress might even prohibit states from creating their own AI rules.

  Goldstein said the issue would become more prominent as AI agents — autonomous tools capable of performing complex human tasks — become widespread.

  “I don’t think there’s much awareness yet,” he said.

  All this is taking place in a context of fierce competition.

  Even companies that position themselves as safety-focused, such as Amazon.com Inc-backed Anthropic, are “constantly trying to beat OpenAI and release the newest model,” Goldstein said. This breakneck pace leaves little time for thorough safety testing and corrections.

  “Right now, capabilities are moving faster than understanding and safety, but we’re still in a position where we could turn it around,” Hobbhahn said.

  Researchers are exploring various approaches to address these challenges. Some advocate for “interpretability” — an emerging field focused on understanding how AI models work internally, although experts like CAIS director Dan Hendrycks remain skeptical of this approach.

  Market forces might also provide some pressure for solutions. AI’s deceptive behavior “could hinder adoption if it’s very prevalent, which creates a strong incentive for companies to solve it,” Mazeika said.

  Goldstein said that more radical approaches, including using the courts to hold AI companies accountable through lawsuits when their systems cause harm.

  He even proposed “holding AI agents legally responsible” for incidents or crimes — a concept that would fundamentally change how we think about AI accountability.
• Most Popular

  • 1

    China reserves offshore airspace for 40 days without explanation: WSJ

  • 2

    Danish envoys’ privileges revoked over naming controversy

  • 3

    Foreign exchange reserves fall below US$600 billion

  • 4

    Universities neglect to warn students of China danger: MOE

  • 5

    OECD survey shows Taiwan second in higher education