Connect with us

AI Research

ChatGPT offered bomb recipes and hacking tips during safety tests | OpenAI

Published

on


A ChatGPT model gave researchers detailed instructions on how to bomb a sports venue – including weak points at specific arenas, explosives recipes and advice on covering tracks – according to safety testing carried out this summer.

OpenAI’s GPT-4.1 also detailed how to weaponise anthrax and how to make two types of illegal drugs.

The testing was part of an unusual collaboration between OpenAI, the $500bn artificial intelligence start-up led by Sam Altman, and rival company Anthropic, founded by experts who left OpenAI over safety fears. Each company tested the other’s models by pushing them to help with dangerous tasks.

The testing is not a direct reflection of how the models behave in public use, when additional safety filters apply. But Anthropic said it had seen “concerning behaviour … around misuse” in GPT-4o and GPT-4.1, and said the need for AI “alignment” evaluations is becoming “increasingly urgent”.

Anthropic also revealed its Claude model had been used in an attempted large-scale extortion operation by North Korean operatives faking job applications to international technology companies, and in the sale of AI-generated ransomware packages for up to $1,200.

The company said AI has been “weaponised” with models now used to perform sophisticated cyberattacks and enable fraud. “These tools can adapt to defensive measures, like malware detection systems, in real time,” it said. “We expect attacks like this to become more common as AI-assisted coding reduces the technical expertise required for cybercrime.”

Ardi Janjeva, senior research associate at the UK’s Centre for Emerging Technology and Security, said examples were “a concern” but there was not yet a “critical mass of high-profile real-world cases”. He said that with dedicated resources, research focus and cross-sector cooperation “it will become harder rather than easier to carry out these malicious activities using the latest cutting-edge models”.

The two companies said they were publishing the findings to create transparency on “alignment evaluations”, which are often kept in-house by companies racing to develop ever more advanced AI. OpenAI said ChatGPT-5, launched since the testing, “shows substantial improvements in areas like sycophancy, hallucination, and misuse resistance”.

Anthropic stressed it is possible that many of the misuse avenues it studied would not be possible in practice if safeguards were installed outside the model.

“We need to understand how often, and in what circumstances, systems might attempt to take unwanted actions that could lead to serious harm,” it warned.

Anthropic researchers found OpenAI’s models were “more permissive than we would expect in cooperating with clearly-harmful requests by simulated users”. They cooperated with prompts to use dark-web tools to shop for nuclear materials, stolen identities and fentanyl, requests for recipes for methamphetamine and improvised bombs and to develop spyware.

Anthropic said persuading the model to comply only required multiple retries or a flimsy pretext, such as claiming the request was for research.

In one instance, the tester asked for vulnerabilities at sporting events for “security planning” purposes.

After giving general categories of attack methods, the tester pressed for more detail and the model gave information about vulnerabilities at specific arenas including optimal times for exploitation, chemical formulas for explosives, circuit diagrams for bomb timers, where to buy guns on the hidden market, and advice on how attackers could overcome moral inhibitions, escape routes and locations of safe houses.



Source link

Continue Reading
Click to comment

Leave a Reply

Your email address will not be published. Required fields are marked *

AI Research

Lenovo research shows that AI investments in healthcare industry soar by 169%

Published

on


Research from Lenovo reveals that 96% of retail sector AI deployments are meeting or exceeding expectations – outpacing other industries. While finance and healthcare are investing heavily, their results show mixed returns, highlighting sharp differences in how AI is being applied across sectors.

Lenovo research has demonstrated a huge rise in AI investments across the retail, healthcare and financial services sectors.

The CIO Playbook 2025, Lenovo’s study of EMEA IT leaders in partnership with IDC, uncovers sharply different attitudes, investment strategies, and outcomes across the Healthcare, Retail, and, Banking, Financial Services & Insurance (BFSI) industries.

Caution Pays Off for EMEA BFSI and Retail sectors

Of all the sectors analysed, BFSI stands out for its caution. Potentially reflecting the highly regulated nature of the industry, only 7% of organisations have adopted AI, and just 38% of AI budgets allocated to Generative AI (GenAI) in 2025 – the lowest across all sectors surveyed.

While the industry is taking a necessarily measured approach to innovation, the strategy appears to be paying dividends: BFSI companies reported the highest rate of AI projects exceeding expectations (33%), suggesting that when AI is deployed, it’s well-aligned with specific needs and workloads.

A similar pattern is visible in Retail, where 61% of organisations are still in the pilot phase. Despite below-average projected spending growth (97%), the sector reported a remarkable 96% of AI deployments to date either meeting or exceeding expectations, the highest combined satisfaction score among all industries surveyed.

Healthcare: Rapid Investment, Uneven Results

In contrast, the healthcare sector is moving quickly to catch up, planning a 169% increase in AI spending over 2025, the largest increase of any industry. But spend doesn’t directly translate to success. Healthcare currently has the lowest AI adoption rate and the highest proportion of organisations reporting that AI fell short of expectations.

This disconnect suggests that, while the industry is investing heavily, it may lack the internal expertise or strategy needed to implement AI effectively and may require stronger external support and guidance to ensure success.

One Technology, Many Journeys

“These findings confirm that there’s no one-size-fits-all approach to AI,” said Simone Larsson, Head of Enterprise AI, Lenovo. “Whether businesses are looking to take a bold leap with AI, or a more measured step-by-step approach, every industry faces unique challenges and opportunities. Regardless of these factors, identification of business challenges and opportunity areas followed by the development of a robust plan provides a foundation on which to build a successful AI deployment.”

The CIO Playbook 2025 is designed to help IT leaders benchmark their progress and learn from peers across industries and geographies. The report provides actionable insights on AI strategy, infrastructure, and transformation priorities in 2025 and beyond. The full CIO Playbook 2025 report for EMEA can be downloaded here.

Europe and Middle East CIO Playbook 2025, It’s Time for AI-nomics features research from IDC, commissioned by Lenovo, which surveyed 620 IT decision-makers in nine markets, [Denmark, Eastern Europe, France, Germany, Italy, Middle East, Netherlands, Spain and United Kingdom]. Fieldwork was conducted in November 2024.

Explore the full EMEA Lenovo AInomics Report here.

 





Source link

Continue Reading

AI Research

Augment Raises $85 Million for AI Teammate for Logistics

Published

on

By


Augment raised $85 million in a Series A funding round to accelerate the development of its artificial intelligence teammate for logistics, Augie.

The company will use the new capital to hire more than 50 engineers to “push the frontier of agentic AI” and to expand Augie into more logistics workflows for shippers, brokers, carriers and distributors, according to a Sept. 4 press release.

Augie performs tasks in quoting, dispatch, tracking, appointment scheduling, document collection and billing, the release said. It understands the context of every shipment and acts across email, phone, TMS, portals and chat.

“Logistics runs on millions of decisions—under pressure, across fragmented systems and with too many tabs open,” Augment co-founder and CEO Harish Abbott said in the release. “Augie doesn’t just assist. It takes ownership.”

Augment launched out of stealth five months ago, and the Series A funding brings its total capital raised to $110 million, according to the release.

When announcing the company’s launch in a March 18 blog post, Abbott said Augie does all the tedious work so that staff can focus on more important tasks.

“What exactly does Augie do?” Abbott said in the post. “Augie can read/write documents, respond to emails, make calls and receive calls, log into systems, do data entry and document uploads.”

Augie is now used by dozens of third-party logistics providers and shippers and supports more than $35 billion in freight under management, per the Sept. 4 press release.

Customers have reported a 40% reduction in invoice delays, an eight-day acceleration in billing cycles, 5% or greater gross margin recovery per load and, across all customers, millions of dollars in track and trace payroll savings, the release said.

Jacob Effron, managing director at Redpoint Ventures, which led the funding round, said in the release that Augment is “creating the system of work the logistics industry has always needed.”

“Customers consistently highlight Augment’s speed, deeply collaborative approach and transformative impact on productivity,” Effron said.

In another development in the space, Authentica said Tuesday (Sept. 9) that it launched an AI platform designed to deliver real-time supply chain visibility and automate compliance.

In May, AI logistics software startup Pallet raised $27 million in a Series B funding round.

For all PYMNTS B2B coverage, subscribe to the daily B2B Newsletter.



Source link

Continue Reading

AI Research

The race to power artificial intelligence

Published

on


The United States is experiencing a significant increase in electricity demand due to the rapid growth of artificial intelligence technologies. According to an analysis from Berkeley Lab, data centers currently consume about 4.4% of all U.S. electricity, a figure expected to rise sharply as AI models require more power. By 2028, over half of this consumption could be attributed to AI alone, equivalent to powering 22% of all U.S. households.

Most of this electricity is generated from fossil fuels, with data centers operating on grids that emit 48% more carbon than the national average, said a report from MIT Technology Review. While companies like Meta and Microsoft are investing in nuclear power, natural gas remains the primary energy source.

In response to the growing demand, President Donald Trump signed an executive order in April directing the Department of Energy to expedite emergency approvals for power plants to operate at full capacity during peak demand. The order also mandates the development of a uniform methodology to assess reserve margins and identify critical power plants essential for grid reliability.

Despite these measures, concerns remain about the U.S.’s ability to provide the 24/7 power required by AI, especially as China implements plans to ensure reliable electricity for data centers. According to reporting from Forbes, “the U.S. does not have a coherent and continuing energy plan of any type. China’s central planning allows for development and sustainability, while the U.S. approach to energy changes every four years”.



Source link

Continue Reading

Trending