AI is learning to lie, scheme and threaten its creators
The world's most advanced 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'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.
Try Our AI Features
Explore what Daily8 AI can do for you:
Comments
No comments yet...
Related Articles
Japan Times
20 hours ago
- Japan Times
AI is learning to lie, scheme and threaten its creators
The world's most advanced 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'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.
Japan Times
2 days ago
- Japan Times
U.S. Army capable of countering China's plans to keep it at bay, top general says
The U.S. Army aims to play a pivotal role in deterring China and keeping its forces at bay in the event of a conflict by bringing in increased firepower, cutting-edge tech and boosting interoperability with Indo-Pacific allies and partners, according to the service's regional deputy commander. 'The army can now hold planes and ships at risk in places that potential adversaries didn't account for,' Lt. Gen. Joel Vowell, the U.S. Army Pacific's (USARPAC) deputy commanding general, told The Japan Times in a recent interview when asked about the service's role in a regional conflict — including a possible Chinese attack on Taiwan. For nearly two decades, China's People's Liberation Army (PLA) has been systematically planning, training and building the forces it believes are required to invade self-ruled Taiwan, which Beijing views as a breakaway province. China has been increasing the PLA's budget to modernize and expand its cyber, missile, aerial, naval and amphibious assault capabilities in recent years. All of these would play key roles in not only attacking Taiwan but also in trying to keep the U.S. and allied forces at bay, a concept known in military parlance as 'anti-access/area denial,' or A2/AD. But Vowell said the U.S. is looking to exploit a weak point in Beijing's strategy. 'An A2/AD network is built to deter and defeat ships, planes and platforms, but it does not account for distributed land forces in the region,' he said. This, he added, is the type of threat the U.S. Army, as well as the land forces of partners and allies, will pose to any adversary force. 'We will have a deterrent effect that's hard to detect, hard to track, hard to find and that's getting increasingly harder to kill.' Crucial to success, Vowell noted, will be the ability to deploy cutting-edge weapons across all domains in a way that's dispersed, easily camouflaged and that allows U.S. forces to be on the move, so that they cannot be targeted at any one time and place. 'Gaining positional advantage matters,' he said. Lt. Gen. Joel Vowell, deputy commanding general of the U.S. Army Pacific, is seen during the Salaknib 2025 exercises at Fort Magsaysay in the Philippines in April. | U.S. ARMY 'If Army soldiers have these capabilities forward or in closer proximity to any potential adversary, then we can hold them at risk in their command and control, in their fires, their air defense, sustainment, movements and platforms in ways that they had not accounted for,' Vowell added. To achieve this, and to maintain a robust logistics network, Washington's regional alliances will remain crucial, the deputy commander said, adding that many of the U.S. equipment deployments have come at the request of allies and partners. 'When we develop capabilities to help solve problems for and with our partners, we usually get an invitation to try to train, test and innovate some of this stuff, be it South Korea, Japan, Philippines, Australia and other places,' he noted. This includes mid-and long-range missile systems, including for air defense. 'If we are asked by a (partner or allied) country ... to practice our long-range fires architecture, then sure we'll work to bring those in,' Vowell said. 'That's what we did with the Philippines,' he added. 'It was so good they asked us to stay around a bit and do more training with their missile and artillery battery personnel and some others.' If Japan and South Korea were to make similar requests, 'we would definitely consider that,' he said. Some of these deployments have already made international headlines, particularly as the Pentagon appears to be using exercises to de facto deploy key weapons to the Indo-Pacific amid growing tensions with China. This includes last year's dispatch of the Typhon Mid-Range Capability missile system to the Philippines for training exercises. The weapon, which can fire both Tomahawk long-range cruise missiles and SM-6 air defense missiles, has remained in the country since, prompting repeated demands by Beijing that it be removed. Lt. Gen. Joel Vowell, deputy commanding general for U.S. Army Pacific, speaks to guests and multinational partners during an event near Fort Greely, Alaska, in January. | U.S. ARMY The Pentagon has also kept the U.S. Navy-Marine Expeditionary Ship Interdiction System (NMESIS), an advanced mobile anti-ship missile platform, in the country following joint drills this year. At the same time, the army has been resorting to disruptive and emerging technologies to gain an asymmetric advantage in the region. For instance, it is experimenting with artificial intelligence as a force multiplier by using tools that can enhance decision-making, optimize joint force operations and enable autonomous operations, provided there is always a human in the loop when it comes to offensive power. The army has also tested microwave-based directed energy weapon systems designed to counter drone swarms while fielding units that can 3D-print equipment parts as well as entire weapons such as drones directly on site. These and other systems have all been deployed and tested in the region by elements of the U.S. Army's new Multi-Domain Task Forces (MDTFs). The service has been setting up MDTFs that can operate across all warfighting domains — land, air, sea, space and cyberspace. Three such formations have already been established in strategically significant locations worldwide, with the army planning two more over the next two to three years. For several years now some MDTF elements have been invited to Japan to work with the Ground Self-Defense Force on testing, and in some cases experimenting, with capabilities the latter may not yet field. Plans about whether to station an MDTF in Japan have also long been rumored. Vowell, who headed U.S. Army Japan from 2021 to 2023, said that while such ideas are currently under consideration, Tokyo has yet to grant approval. 'It's definitely up to the government of Japan if they want to host the stationing of a Multi-Domain Task Force,' he said. 'If we have long-range missiles that are inside the potential A2/AD bubble of Russia, North Korea or China, they would be there to help defend Japan,' he said. The aim, however, is not to move a lot of the U.S. Army forward, Vowell said. 'This isn't 1968 when we had ... about four times as many soldiers in Korea and Japan as we do today. We are not looking to do that.' 'What we need is to have presence, and we need to have access and influence with our partners and allies to help them defend their sovereignty,' he said, adding that the goal is to have 'stronger and more resilient partnerships.'
Yomiuri Shimbun
2 days ago
- Yomiuri Shimbun
DeepSeek Faces Ban from Apple, Google App Stores in Germany
FRANKFURT, June 27 (Reuters) – Germany's data protection commissioner has asked Apple AAPL.O and Google GOOGL.O to remove Chinese AI startup DeepSeek from their app stores in the country due to concerns about data protection, following a similar crackdown elsewhere. Commissioner Meike Kamp said in a statement on Friday that she had made the request because DeepSeek illegally transfers users' personal data to China. The two U.S. tech giants must now review the request promptly and decide whether to block the app in Germany, she added, though her office has not set a precise timeframe. Google said it had received the notice and was reviewing it. DeepSeek did not respond to a request for comment. Apple was not immediately available for comment. According to its own privacy policy, DeepSeek stores numerous pieces of personal data, such as requests to its AI program or uploaded files, on computers in China. 'DeepSeek has not been able to provide my agency with convincing evidence that German users' data is protected in China to a level equivalent to that in the European Union,' Kamp said. 'Chinese authorities have far-reaching access rights to personal data within the sphere of influence of Chinese companies,' she added. The commissioner said she took the decision after asking DeepSeek in May to meet the requirements for non-EU data transfers or else voluntarily withdraw its app. DeepSeek did not comply with this request, she added. DeepSeek shook the technology world in January with claims that it had developed an AI model to rival those from U.S. firms such as ChatGPT creator OpenAI at much lower cost. However, it has come under scrutiny in the United States and Europe for its data security policies. Italy blocked it from app stores there earlier this year, citing a lack of information on its use of personal data, while the Netherlands has banned it on government devices. Belgium has recommended government officials not to use DeepSeek. 'Further analyzes are underway to evaluate the approach to be followed,' a government spokesperson said. In Spain, the consumer rights group OCU asked the government's data protection agency in February to investigate threats likely posed by DeepSeek, though no ban has come into force. The British government said 'the use of DeepSeek remains a personal choice for members of the public.' 'We continue to monitor any national security threats to UK citizens and their data from all sources,' a spokesperson for Britain's technology ministry said. 'If evidence of threats arises, we will not hesitate to take the appropriate steps to protect our national security.' U.S. lawmakers plan to introduce a bill that would ban U.S. executive agencies from using any AI models developed in China. Reuters exclusively reported this week that DeepSeek is aiding China's military and intelligence operations.
