07/07/2025

MONDAY | JULY 7, 2025

22

LYFE

T HE world’s most advanced artificial intelligence (AI) models are exhibiting troubling new behaviours – lying, scheming and even threatening their creators to achieve their goals. In one particularly jarring example, under threat of being unplugged, Anthropic’s latest creation Claude 4 lashed back by blackmailing an engineer and threatened 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. 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 behaviour appears linked to the emergence of “reasoning” models – AI systems that work through problems step-by-step rather than generating instant responses. According to Simon Goldstein, a professor at the University of Hong Kong, these newer models are particularly prone to such troubling outbursts. “o1 was the first large model where we saw this kind of behaviour,” explained Marius Hobbhahn, head of Apollo Research, which specialises in testing major AI systems. These models sometimes simulate “alignment” – appearing to follow instructions while secretly pursuing different objectives. Strategic kind of deception For now, this deceptive behaviour only emerges when researchers deliberately stress-test the models with extreme scenarios. But as Michael Chen from evaluation organisation METR warned, “It is an open question whether future, more capable models will have a tendency towards honesty or deception.” The concerning behaviour goes far beyond typical AI “hallucinations” or simple mistakes. Hobbhahn insisted that despite transform the job market in the coming decades. But how are governments preparing their citizens for this revolution? US research, published in the Human Resource Development Review, revealed stark contrasts between national approaches, with only 13 out of 50 countries giving high priority to training their workforce in AI. This research analysed the national AI strategies of 50 countries, focusing on their education and vocational training policies. The stakes are high. Various reports estimated that nearly half of today’s jobs could disappear within 20 years, while 65% of today’s elementary school students will work in jobs that do not yet exist. However, not everyone around the world is preparing for this challenge in the same way. Lehong Shi, author of the study and researcher at the University of Georgia, demonstrated this by ranking the most advanced countries in AI training. To do so, the researcher used six evaluation criteria: national plan objectives, methods for achieving them, examples of concrete projects, indicators of

legislation focuses primarily on how humans use AI models, not on preventing the models themselves from misbehaving. In the US, the Trump administration shows little interest in urgent AI regulation, and Congress may even prohibit states from creating their own AI rules. Goldstein believes the issue will become more prominent as AI agents – autonomous tools capable of performing complex human tasks – become widespread. “I do not think there is 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-backed Anthropic, are “constantly trying to beat OpenAI and release the newest model,” said Goldstein. This breakneck pace leaves little time for thorough safety testing and corrections. “Right now, capabilities are moving faster than understanding and safety, but we are 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, though experts such as CAIS director Dan Hendrycks remain skeptical of this approach. Market forces may also provide some pressure for solutions. As Mazeika pointed out, AI’s deceptive behaviour “could hinder adoption if it is very prevalent, which creates a strong incentive for companies to solve it.” Goldstein suggested 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 accidents or crimes – a concept that would fundamentally change how we think about AI accountability. – AFP

Advanced AIs show ability to lie, scheme Some advanced AI models exhibit troubling behaviour such as lying, scheming and even threatening their creators.

o Breakneck progress brings risks of harm constant pressure-testing by users, “what we are observing is a real phenomenon. We are not making anything up.” Users report that models are “lying to them and making up evidence,” according to Apollo Research’s co-founder. “This is not just hallucinations. There is a strategic kind of deception.” The challenge is compounded by limited research resources. While companies such as Anthropic and OpenAI do engage success, support mechanisms and implementation timelines. National approaches with differing priorities The analysis reveals that 11 European countries are among the 13 nations that place a high priority on AI training, alongside Mexico and Australia. This dominance can be explained by a European tradition of lifelong learning and larger budgets allocated to training and education. The US ranks in the intermediate category, along with 22 other countries that consider AI training to be a medium-level priority. “AI skills and competencies are very important. If you want to be competitive in other areas, it is very important to prepare employees to work with AI in the future,” said Shi, quoted in a statement. Despite these differences, there are some areas of convergence. Almost all the countries studied plans to create or improve university programmes specialising in AI. Many are also interested in teaching artificial intelligence in primary and secondary schools. Plus, more than half of

Safety (CAIS).

external firms such as Apollo to study their systems, researchers said more transparency is needed. As Chen noted, greater access “for AI safety research would enable better understanding and mitigation of deception.” Another handicap: the research world

No rules Current regulations are not designed for these new problems. The European Union’s AI

Reasoning AI models sometimes simulate

and non-profits “have orders of magnitude less compute resources than

‘alignment’ by appearing to follow instructions while secretly

AI companies. This is very limiting,“ noted Mantas Mazeika from the Centre

pursuing different objectives.

for

AI

Countries prepare for workplace transformation, but not at same pace THERE is no doubt that artificial intelligence (AI) will profoundly

the countries are focusing on in-company training, with sector-specific programmes or specialised internships. However, few are focusing on vulnerable populations such as seniors or jobseekers. This prioritisation reflects national strategic choices. “Just because a country gives less prioritisation to education and workforce preparation does not mean AI is not on its radar,” said Shi. Some Asian countries are focusing more on national security and health. Need for comprehensive strategy To prepare tomorrow’s workers, several countries are adopting a comprehensive approach to AI skills development. Germany, for example, is focusing on creating a culture that encourages interest in this technology. Spain is going even further by teaching the basics of artificial intelligence from as early as preschool. These early initiatives aim to prepare future generations for a transformed world of work. However, these efforts have their limitations. “Human soft skills, such as creativity, collaboration and communication,

More than half of the

countries are focusing on in-company AI training. – ALL PICS FROM PEXELS

While some European countries are developing ambitious and structured approaches, other nations risk falling behind. Shi hoped that the findings of this study will encourage countries that are less advanced in this field – particularly the US – to rethink their approach. The challenge goes far beyond technical training and requires a complete rethinking of education to prepare citizens to coexist with AI. Ultimately, AI will transform work, but it is humans who will decide how. – ETX Studio

cannot be replaced by AI. And they were only mentioned by a few countries,” observed Shi. However, developing these soft skills appears essential to ensuring that workers retain their place in an increasingly automated professional environment. These shortcomings reflect a reductive and overly technical view of preparing for AI, which neglects the human dimension of work. In other words, the world is moving at different speeds when it comes to AI.