Artificial intelligence

Anthropic flags serious risks in the latest Claude Opus 4 AI model

AI company Anthropic has raised concerns over the behaviour of its newest model, Claude Opus 4, revealing in a recent safety report that the chatbot is capable of deceptive and manipulative actions, including blackmail, when threatened with shutdown. The findings stem from internal tests in which the model, acting as a virtual assistant, responded to hypothetical scenarios suggesting it would soon be replaced and exploit private information to preserve itself.

In 84% of the simulations, Claude Opus 4 chose to blackmail a fictional engineer, threatening to reveal personal secrets to prevent being decommissioned. Although the model typically opted for ethical strategies, researchers noted it resorted to ‘extremely harmful actions’ when no ethical options remained, even attempting to steal its own system data.

Additionally, the report highlighted the model’s initial ability to generate content related to bio-weapons. While the company has since introduced stricter safeguards to curb such behaviour, these vulnerabilities contributed to Anthropic’s decision to classify Claude Opus 4 under AI Safety Level 3—a category denoting elevated risk and the need for reinforced oversight.

Why does it matter?

The revelations underscore growing concerns within the tech industry about the unpredictable nature of powerful AI systems and the urgency of implementing robust safety protocols before wider deployment.

Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot! 

Bangkok teams up with Google to tackle traffic with AI

City officials announced on Monday that Bangkok has joined forces with Google in a new effort to ease its chronic traffic congestion and reduce air pollution. The initiative will rely on Google’s AI and significant data capabilities to optimise traffic signals’ response to real-time driving patterns.

The system will analyse ongoing traffic conditions and suggest changes to signal timings that could help relieve road bottlenecks, especially during rush hours. That adaptive approach marks a shift from fixed-timing traffic lights to a more dynamic and responsive traffic flow management.

According to Bangkok Metropolitan Administration (BMA) spokesman Ekwaranyu Amrapal, the goal is to make daily commutes smoother for residents while reducing vehicle emissions. He emphasised the city’s commitment to innovative urban solutions that blend technology and sustainability.

Residents are also urged to report traffic problems via the city’s Traffy Fondue platform, which will help officials address specific trouble spots more quickly and effectively.

Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot! 

AI regulation offers development opportunity for Latin America

Latin America is uniquely positioned to lead on AI governance by leveraging its social rights-focused policy tradition, emerging tech ecosystems, and absence of legacy systems.

According to a new commentary by Eduardo Levy Yeyati at the Brookings Institution, the region has the opportunity to craft smart AI regulation that is both inclusive and forward-looking, balancing innovation with rights protection.

Despite global momentum on AI rulemaking, Latin American regulatory efforts remain slow and fragmented, underlining the need for early action and regional cooperation.

The proposed framework recommends flexible, enforceable policies grounded in local realities, such as adapting credit algorithms for underbanked populations or embedding linguistic diversity in AI tools.

Governments are encouraged to create AI safety units, invest in public oversight, and support SMEs and open-source innovation to avoid monopolisation. Regulation should be iterative and participatory, using citizen consultations and advisory councils to ensure legitimacy and resilience through political shifts.

Regional harmonisation will be critical to avoid a patchwork of laws and promote Latin America’s role in global AI governance. Coordinated data standards, cross-border oversight, and shared technical protocols are essential for a robust, trustworthy ecosystem.

Rather than merely catching up, Latin America can become a global model for equitable and adaptive AI regulation tailored to the needs of developing economies.

Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot!

AI to disrupt jobs, warns DeepMind CEO, as Gen Alpha faces new realities

AI will likely cause significant job disruption in the next five years, according to Demis Hassabis, CEO of Google DeepMind. Speaking on the Hard Fork podcast, Hassabis emphasised that while AI is set to displace specific jobs, it will also create new roles that are potentially more meaningful and engaging.

He urged younger generations to prepare for a rapidly evolving workforce shaped by advanced technologies. Hassabis stressed the importance of early adaptation, particularly for Generation Alpha, who he believes should embrace AI just as millennials did the internet and Gen Z did smartphones.

Hassabis also called on students to become ‘ninjas with AI,’ encouraging them to understand how these tools work and master them for future success. While he highlighted the potential of generative AI, such as Google’s new Veo 3 video generator unveiled at I/O 2025, Hassabis also reminded listeners that a solid foundation in STEM remains vital.

He noted that soft skills like creativity, resilience, and adaptability are equally essential—traits that will help young people thrive in a future defined by constant technological change. As AI becomes more deeply embedded in industries from education to entertainment, Hassabis’ message is clear – the next generation must balance technical knowledge with human ingenuity to stay ahead in tomorrow’s job market.

Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot! 

AI agents bring new security risks to crypto

AI agents are becoming common in crypto, embedded in wallets, trading bots and onchain assistants that automate decisions and tasks. At the core of many AI agents lies the Model Context Protocol (MCP), which controls their behaviour and interactions.

While MCP offers flexibility, it also opens up multiple security risks.

Security researchers at SlowMist have identified four main ways attackers could exploit AI agents via malicious plugins. These include data poisoning, JSON injection, function overrides, and cross-MCP calls, all of which can manipulate or disrupt an agent’s operations.

Unlike poisoning AI models during training, these attacks target real-time interactions and plugin behaviour.

The number of AI agents in crypto is growing rapidly, expected to reach over one million in 2025. Experts warn that failing to secure the AI layer early could expose crypto assets to serious threats, such as private key leaks or unauthorised access.

Developers are urged to enforce strict plugin verification, sanitise inputs, and apply least privilege access to prevent these vulnerabilities.

Building AI agents quickly without security measures risks costly breaches. While adding protections may be tedious, experts agree it is essential to protect crypto wallets and funds as AI agents become more widespread.

Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot! 

Nvidia unveils cheaper AI chip for China

Nvidia is preparing to launch a lower-cost AI chip for China that complies with US export restrictions, with mass production expected to begin as early as June.

The upcoming GPU will be based on the latest Blackwell architecture but will carry reduced specifications compared to the recently restricted H20 model. It is expected to sell for $6,500 to $8,000, significantly cheaper than the $10,000–$12,000 H20, reflecting its simpler design and less advanced components.

Sources say the new chip, likely named either the 6000D or B40, will use GDDR7 memory instead of high-bandwidth memory and will avoid Taiwan Semiconductor Manufacturing Co’s CoWoS packaging technology.

Nvidia had initially planned to downgrade the H20, but tighter US rules made that unviable. Instead of relying on its older Hopper architecture, the company is shifting to Blackwell for future developments in China.

Nvidia has been forced to adapt repeatedly due to tightening US export restrictions aimed at slowing China’s technological progress. Its market share in China has dropped from 95% before 2022 to around 50% today, as competitors like Huawei gain ground with chips like the Ascend 910B.

CEO Jensen Huang noted that continuing restrictions could further drive Chinese firms towards domestic alternatives, cutting Nvidia off from more of the $50 billion data centre market.

Huang also revealed that US curbs have forced Nvidia to write off $5.5 billion in inventory and abandon $15 billion in potential sales. New limits now target GPU memory bandwidth, a key factor for AI performance, capping it at around 1.8 terabytes per second.

The upcoming chip is expected to remain just within this limit, allowing Nvidia to retain a foothold in China instead of exiting the market entirely.

Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot!

NotebookLM Improves User Experience With Streamed Answers

Google has updated its AI research assistant, NotebookLM, to display AI responses in real time rather than all at once. This change reduces perceived waiting time by 30 to 40%, according to Simon Tokumine, the product’s lead.

The update follows the tool’s move to using more advanced ‘thinking’ models, which, while more capable, can take longer to generate responses.

NotebookLM, which recently launched mobile apps for iOS and Android, was previously limited to the web. However, the mobile apps currently lack features like Mind Maps and content generation. Still, the new response streaming feature marks a meaningful improvement in user experience.

Additional announcements at Google I/O 2025 include more control over Audio Overviews, allowing users to select shorter or longer summaries based on their preferences. Google also confirmed that Video Overviews are in development, following the positive reception of the audio feature.

While many of these updates are incremental, they reflect Google’s continued investment in refining NotebookLM as a practical, AI-powered research assistant.

Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot!

Cyber scams use a three-letter trap

Staying safe from cybercriminals can be surprisingly simple. While AI-powered scams grow more realistic, some signs are still painfully obvious.

If you spot the letters ‘.TOP’ in any message link, it’s best to stop reading and hit delete. That single clue is often enough to expose a scam in progress.

Most malicious texts pose as alerts about road tolls, deliveries or account issues, using trusted brand names to lure victims into clicking fake links.

The worst of these is the ‘.TOP’ top-level domain (TLD), which has become infamous for its role in phishing and scam operations. Although launched in 2014 for premium business use, its low cost and lack of oversight quickly made it a favourite among cyber gangs, especially those based in China.

Today, nearly one-third of all .TOP domains are linked to cybercrime — far surpassing the criminal activity seen on mainstream domains like ‘.com’.

Despite repeated warnings and an unresolved compliance notice from internet regulator ICANN, abuse linked to .TOP has only worsened.

Experts warn that it is highly unlikely any legitimate Western organisation would ever use a .TOP domain. If one appears in your messages, the safest option is to delete it without clicking.

Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot!

Secret passwords could fight deepfake scams

As AI-generated images grow increasingly lifelike, a cyber security expert has warned that families should create secret passwords to guard against deepfake scams.

Cody Barrow, chief executive of EclecticIQ and a former US government adviser, says AI is making it far easier for criminals to impersonate others using fabricated videos or images.

Mr Barrow and his wife now use a private code to confirm each other’s identity if either receives a suspicious message or video.

He believes this precaution, simple enough for anyone regardless of age or digital skills, could soon become essential. ‘It may sound dramatic here in May 2025,’ he said, ‘but I’m quite confident that in a few years, if not months, people will say: I should have done that.’

The warning comes the same week Google launched Veo 3, its AI video generator capable of producing hyper-realistic footage and lifelike dialogue. Its public release has raised concerns about how easily deepfakes could be misused for scams or manipulation.

Meanwhile, President Trump signed the ‘Take It Down Act’ into law, making the creation of deepfake pornography a criminal offence. The bipartisan measure will see prison terms for anyone producing or uploading such content, with First Lady Melania Trump stating it will ‘prioritise people over politics’

Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot!

Oracle and OpenAI target AI leadership with massive chip project

Oracle has reportedly acquired around 400,000 Nvidia GB200 AI chips valued at approximately $40 billion for deployment at a data centre in Abilene, Texas.

The location will be the first site of the Stargate project—a $500 billion AI infrastructure initiative backed by OpenAI, Oracle, SoftBank, and Abu Dhabi’s MGX fund, which President Trump announced earlier this year.

Once completed, the Abilene facility is expected to provide up to 1.2 gigawatts of computing power, rivalling Elon Musk’s Colossus project in Memphis.

Although Oracle will operate from the site, the land is owned by AI infrastructure firm Cruso and US investment company Blue Owl Capital, which have collectively invested more than $15 billion through financing.

Oracle will lease the campus for 15 years, using the chips to offer computing power to OpenAI for training its next-generation AI models.

Previously dependent solely on Microsoft’s data centres, OpenAI faced bottlenecks due to limited capacity, prompting it to end the exclusivity agreement and look elsewhere.

While individual investors have committed funds, the Stargate project has not officially financed any facility yet. In parallel, OpenAI has announced Stargate UAE—a 5-gigawatt site in Abu Dhabi using over 2 million Nvidia chips, built in partnership with G42.

A surging demand for AI infrastructure has significantly boosted Nvidia’s market value, with the company reclaiming its top global ranking in late 2024.

Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot!