Economic

Musk’s chatbot Grok removes offensive content

Elon Musk’s AI chatbot Grok has removed several controversial posts after they were flagged as anti-Semitic and accused of praising Adolf Hitler.

The deletions followed backlash from users on X and criticism from the Anti-Defamation League (ADL), which condemned the language as dangerous and extremist.

Grok, developed by Musk’s xAI company, sparked outrage after stating Hitler would be well-suited to tackle anti-White hatred and claiming he would ‘handle it decisively’. The chatbot also made troubling comments about Jewish surnames and referred to Hitler as ‘history’s moustache man’.

In response, xAI acknowledged the issue and said it had begun filtering out hate speech before posts go live. The company credited user feedback for helping identify weaknesses in Grok’s training data and pledged ongoing updates to improve the model’s accuracy.

The ADL criticised the chatbot’s behaviour as ‘irresponsible’ and warned that such AI-generated rhetoric fuels rising anti-Semitism online.

It is not the first time Grok has been caught in controversy — earlier this year, the bot repeated White genocide conspiracy theories, which xAI blamed on an unauthorised software change.

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

ChatGPT quietly tests new ‘Study Together’ feature for education

A few ChatGPT users have noticed a new option called ‘Study Together’ appearing among available tools, though OpenAI has yet to confirm any official rollout. The feature seems designed to make ChatGPT a more interactive educational companion than just delivering instant answers.

Rather than offering direct solutions, the tool prompts users to think for themselves by asking questions, potentially turning ChatGPT into a digital tutor.

Some speculate the mode might eventually allow multiple users to study together in real-time, mimicking a virtual study group environment.

With the chatbot already playing a significant role in classrooms — helping teachers plan lessons or assisting students with homework — the ‘Study Together’ feature might help guide users toward deeper learning instead of enabling shortcuts.

Critics have warned that AI tools like ChatGPT risk undermining education, so it could be a strategic shift to encourage more constructive academic use.

OpenAI has not confirmed when or if the feature will launch publicly, or whether it will be limited to ChatGPT Plus users. When asked, ChatGPT only replied that nothing had been officially announced.

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

Sam Altman shrugs off Meta poaching, backs Trump, jabs at Musk

OpenAI CEO Sam Altman addressed multiple hot topics during the Sun Valley conference, including Meta’s aggressive recruitment of top AI researchers, his strained relationship with Elon Musk, and a surprising show of support for Donald Trump.

Altman downplayed Meta’s talent raids, saying he had not spoken to Mark Zuckerberg since the Meta CEO lured away three OpenAI researchers with a $100 million signing bonus. All three had worked at OpenAI’s Zurich office, which opened in 2024.

Despite the losses, Altman described the situation as ‘fine’ and ‘good’, suggesting OpenAI’s mission continues to retain top talent.

The OpenAI chief also took a subtle swipe at Meta’s smartglasses, saying he doesn’t like wearable tech and implying his company has no plans to follow suit.

On the topic of Elon Musk, Altman laughed off their rivalry, saying only that Musk’s bust-ups with everybody, and hinting at the long-running tension between the two former co-founders.

Perhaps most notably, Altman expressed disillusionment with the Democratic Party, saying he no longer feels represented by mainstream figures he once supported.

He praised Donald Trump’s focus on AI infrastructure. He even donated $1 million to Trump’s inaugural fund — a gesture reflecting a broader shift among Silicon Valley leaders warming to Trump as his popularity rises.

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

Cybercrime soars as firms underfund defences

Nearly four in ten UK businesses (38 %) do not allocate a dedicated cybersecurity budget, even as cybercrime costs hit an estimated £64 billion over three years.

Smaller enterprises are particularly vulnerable, with 15 % reporting breaches linked to underfunding.

Almost half of organisations (45 %) rely solely on in‑house defences, with only 8 % securing standalone cyber insurance, exposing many to evolving threats.

Common attacks include phishing campaigns, AI‑powered malware and DDoS, yet cybersecurity typically receives just 11 % of IT budgets.

Security professionals call for stronger board‑level involvement and increased collaboration with specialists and regulators.

They caution that businesses risk suffering further financial and reputational damage without proactive budgeting and external expertise.

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

Phishing 2.0: How AI is making cyber scams more convincing

Phishing remains among the most widespread and dangerous cyber threats, especially for individuals and small businesses. These attacks rely on deception—emails, texts, or social messages that impersonate trusted sources to trick people into giving up sensitive information.

Cybercriminals exploit urgency and fear. A typical example is a fake email from a bank saying your account is at risk, prompting you to click a malicious link. Even when emails look legitimate, subtle details—like a strange sender address—can be red flags.

In one recent scam, Netflix users received fake alerts about payment failures. The link led to a fake login page where credentials and payment data were stolen. Similar tactics have been used against QuickBooks users, small businesses, and Microsoft 365 customers.

Small businesses are frequent targets due to limited security resources. Emails mimicking vendors or tech companies often trick employees into handing over credentials, giving attackers access to sensitive systems.

Phishing works because it preys on human psychology: trust, fear, and urgency. And with AI, attackers can now generate more convincing content, making detection harder than ever.

Protection starts with vigilance. Always check sender addresses, avoid clicking suspicious links, and enable multi-factor authentication (MFA). Employee training, secure protocols for sensitive requests, and phishing simulations are critical for businesses.

Phishing attacks will continue to grow in sophistication, but with awareness and layered security practices, users and businesses can stay ahead of the threat.

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

Lithuania questions legality of Robinhood’s new tokens

Robinhood’s new blockchain tokens, which are tied to firms like SpaceX and OpenAI, are under EU scrutiny, with Lithuania’s central bank reviewing whether the product meets financial rules.

The tokens, introduced on 30 July, allow retail investors to gain exposure to high-profile private firms through digital assets. Although Robinhood offered a promotional giveaway to attract EU users, questions quickly arose over the product’s legal classification and how it was marketed to the public.

OpenAI has publicly stated it has no affiliation with Robinhood and has not authorised share transfers. Robinhood responded by confirming that the tokens do not represent actual equity but provide access to the value of private firms via non-equity instruments.

Regulators are now assessing whether the product meets compliance standards and whether investor information has been presented clearly and accurately. The central bank has requested further details before issuing a formal judgement.

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

How agentic AI is transforming cybersecurity

Cybersecurity is gaining a new teammate—one that never sleeps and acts independently. Agentic AI doesn’t wait for instructions. It detects threats, investigates, and responds in real-time. This new class of AI is beginning to change the way we approach cyber defence.

Unlike traditional AI systems, Agentic AI operates with autonomy. It sets objectives, adapts to environments, and self-corrects without waiting for human input. In cybersecurity, this means instant detection and response, beyond simple automation.

With networks more complex than ever, security teams are stretched thin. Agentic AI offers relief by executing actions like isolating compromised systems or rewriting firewall rules. This technology promises to ease alert fatigue and keep up with evasive threats.

A 2025 Deloitte report says 25% of GenAI-using firms will pilot Agentic AI this year. SailPoint found that 98% of organisations will expand AI agent use in the next 12 months. But rapid adoption also raises concern—96% of tech workers see AI agents as security risks.

The integration of AI agents is expanding to cloud, endpoints, and even physical security. Yet with new power comes new vulnerabilities—from adversaries mimicking AI behaviour to the risk of excessive automation without human checks.

Key challenges include ethical bias, unpredictable errors, and uncertain regulation. In sectors like healthcare and finance, oversight and governance must keep pace. The solution lies in balanced control and continuous human-AI collaboration.

Cybersecurity careers are shifting in response. Hybrid roles such as AI Security Analysts and Threat Intelligence Automation Architects are emerging. To stay relevant, professionals must bridge AI knowledge with security architecture.

Agentic AI is redefining cybersecurity. It boosts speed and intelligence but demands new skills and strong leadership. Adaptation is essential for those who wish to thrive in tomorrow’s AI-driven security landscape.

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

US targets Southeast Asia to stop AI chip leaks to China

The US is preparing stricter export controls on high-end Nvidia AI chips destined for Malaysia and Thailand, in a move to block China’s indirect access to advanced GPU hardware.

According to sources cited by Bloomberg, the new restrictions would require exporters to obtain licences before sending AI processors to either country.

The change follows reports that Chinese engineers have hand-carried data to Malaysia for AI training after Singapore began restricting chip re-exports.

Washington suspects Chinese firms are using Southeast Asian intermediaries, including shell companies, to bypass existing export bans on AI chips like Nvidia’s H100.

Although some easing has occurred between the US and China in areas such as ethane and engine components, Washington remains committed to its broader decoupling strategy. The proposed measures will reportedly include safeguards to prevent regional supply chain disruption.

Malaysia’s Trade Minister confirmed earlier this year that the US had requested detailed monitoring of all Nvidia chip shipments into the country.

As the global race for AI dominance intensifies, Washington appears determined to tighten enforcement and limit Beijing’s access to advanced computing power.

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

Ransomware disrupts Ingram Micro’s systems and operations

Ingram Micro has confirmed a ransomware attack that affected internal systems and forced some services offline. The global IT distributor says it acted quickly to contain the incident, implemented mitigation steps, and involved cybersecurity experts.

The company is working with a third-party firm to investigate the breach and has informed law enforcement. Order processing and shipping operations have been disrupted while systems are being restored.

While details remain limited, the attack is reportedly linked to the SafePay ransomware group.

According to BleepingComputer, the gang exploited Ingram’s GlobalProtect VPN to gain access last Thursday.

In response, Ingram Micro shut down multiple platforms, including GlobalProtect VPN and its Xvantage AI platform. Employees were instructed to work remotely as a precaution during the response effort.

SafePay first appeared in late 2024 and has targeted over 220 companies. It often breaches networks using password spraying and compromised credentials, primarily through VPNs.

Ingram Micro has not disclosed what data was accessed or the size of the ransom demand.

The company apologised for the disruption and said it is working to restore systems as quickly as possible.

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

Pakistan launches AI customs system to tackle tax evasion

Pakistan has launched its first AI-powered Customs Clearance and Risk Management System (RMS) to cut tax evasion, reduce corruption, and modernise port operations by automating inspections and declarations.

The initiative, part of broader digital reforms, is led by the Federal Board of Revenue (FBR) with support from the Intelligence Bureau.

By minimising human involvement in customs procedures, the system enables faster, fairer, and more transparent processing. It uses AI and automated bots to assess goods’ value and classification, improve risk profiling, and streamline green channel clearances.

Early trials showed a 92% boost in system performance and more than double the efficiency in identifying compliant cargo.

Prime Minister Shehbaz Sharif praised the collaboration between the FBR and IB, calling the initiative a key pillar of national economic reform. He urged full integration of the system into the country’s digital infrastructure and reaffirmed tax reform as a government priority.

The AI system is also expected to close loopholes in under-invoicing and misdeclaration, which have long been used to avoid duties.

Meanwhile, video analytics technology is trialled to detect factory tax fraud, with early tests showing 98% accuracy. In recent enforcement efforts, authorities recovered Rs178 billion, highlighting the potential of data-driven approaches in tackling fiscal losses.

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