xAI unveils Grok 4 with top benchmark scores

Elon Musk’s AI company, xAI, has launched its latest flagship model, Grok 4, alongside an ultra-premium $300 monthly plan named SuperGrok Heavy.

Grok 4, which competes with OpenAI’s ChatGPT and Google’s Gemini, can handle complex queries and interpret images. It is now integrated more deeply into the social media platform X, which Musk also owns.

Despite recent controversy, including antisemitic responses generated by Grok’s official X account, xAI focused on showcasing the model’s performance.

Musk claimed Grok 4 is ‘better than PhD level’ in all academic subjects and revealed a high-performing version called Grok 4 Heavy, which uses multiple AI agents to solve problems collaboratively.

The models scored strongly on benchmark exams, including a 25.4% score for Grok 4 on Humanity’s Last Exam, outperforming major rivals. With tools enabled, Grok 4 Heavy reached 44.4%, nearly doubling OpenAI’s and Google’s results.

It also achieved a leading score of 16.2% on the ARC-AGI-2 pattern recognition test, nearly double that of Claude Opus 4.

xAI is targeting developers through its API and enterprise partnerships while teasing upcoming tools: an AI coding model in August, a multi-modal agent in September, and video generation in October.

Yet the road ahead may be rocky, as the company works to overcome trust issues and position Grok as a serious rival in the AI arms race.

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

AI fluency is the new office software skill

As tools like ChatGPT, Copilot, and other generative AI systems become embedded in daily workflows, employers increasingly prioritise a new skill: AI fluency.

Much like proficiency in office software became essential in the past, knowing how to collaborate effectively with AI is now a growing requirement across industries.

But interacting with AI isn’t always intuitive. Many users encounter generic or unhelpful responses from chatbots and assume the technology is limited. In reality, AI systems rely heavily on the context they are given, and that’s where users come in.

Rather than considering AI as a search engine, it helps to see it as a partner needing guidance. A vague prompt like ‘write a proposal’ is unlikely to produce meaningful results. A better approach provides background, direction, and clear expectations.

One practical framework is CATS: context, angle, task, and style.

Context sets the stage. It includes your role, the situation, the audience, and constraints. For example, ‘I’m a nonprofit director writing a grant proposal for an environmental education program in urban schools’ offers much more to work with than a general request.

Angle defines the perspective. You can ask the AI to act as a peer reviewer, a mentor, or even a sceptical audience member. The roles help shape the tone and focus of the response.

Task clarifies the action you want. Instead of asking for help with a presentation, try ‘Suggest three ways to improve my opening slide for an audience of small business owners.’

Style determines the format and tone. Whether you need a formal report, a friendly email, or an outline in bullet points, specifying the style helps the AI deliver a more relevant output.

Beyond prompts, users can also practice context engineering—managing the environment around the prompt. The method includes uploading relevant documents, building on previous chats, or setting parameters through instructions. The steps help tailor responses more closely to your needs.

Think of prompting as a conversation, not a one-shot command. If the initial response isn’t ideal, clarify, refine, or build on it. Ask follow-up questions, adjust your instructions, or extract functional elements to develop further in a new thread.

That said, it’s essential to stay critical. AI systems can mimic natural conversation, but don’t truly understand the information they provide. Human oversight remains crucial. Always verify outputs, especially in professional or high-stakes contexts.

Ultimately, AI tools are powerful collaborators—but only when paired with clear guidance and human judgment. Provide the correct input, and you’ll often find the output exceeds expectations.

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

Privacy concerns rise over Gemini’s on‑device data access

From 7 July 2025, Google’s Gemini AI will default to accessing your WhatsApp, SMS and call apps, even without Gemini Apps Activity enabled, through an Android OS’ System Intelligence’ integration.

Google insists the assistant cannot read or summarise your WhatsApp messages; it only performs actions like sending replies and accessing notifications.

Integration occurs at the operating‑system level, granting Gemini enhanced control over third‑party apps, including reading and responding to notifications or handling media.

However, this has prompted criticism from privacy‑minded users, who view it as intrusive data access, even though Google maintains no off‑device content sharing.

Alarmed users quickly turned off the feature via Gemini’s in‑app settings or resorted to more advanced measures, like removing Gemini with ADB or turning off the Google app entirely.

The controversy highlights growing concerns over how deeply OS‑level AI tools can access personal data, blurring the lines between convenience and privacy.

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

LG’s Exaone Path 2.0 uses AI to transform genetic testing

LG AI Research has introduced Exaone Path 2.0, an upgraded AI model designed to analyse pathology images for disease diagnosis, significantly reducing the time required for genetic testing.

The new model, unveiled Wednesday, can reportedly process pathology images in under a minute—a significant shift from conventional genetic testing methods that often take more than two weeks.

According to LG, the AI system offers enhanced accuracy in detecting genetic mutations and gene expression patterns by learning from detailed image patches and full-slide pathology data.

Developed by LG AI Research, a division of the LG Group, Exaone Path 2.0 is trained on over 10,000 whole-slide images (WSIs) and multiomics pairs, enabling it to integrate structural information with molecular biology insights. The company said it has achieved a 78.4 percent accuracy rate in predicting genetic mutations.

The model has also been tailored for specific applications in oncology, including lung and colorectal cancers, where it can help clinicians identify patient groups most likely to benefit from targeted therapies.

LG AI Research is collaborating with Professor Hwang Tae-hyun and his team at Vanderbilt University Medical Centre in the US to further its application in real-world clinical settings.

Their shared goal is to develop a multimodal medical AI platform that can support precision medicine directly within clinical environments.

Hwang, a key contributor to the US government’s Cancer Moonshot program and founder of the Molecular AI Initiative at Vanderbilt, emphasised that the aim is to create AI tools usable by clinicians in active medical practice, rather than limiting innovation to the lab.

In addition to oncology, LG AI Research plans to extend its multimodal AI initiatives into transplant rejection, immunology, and diabetes.

It is also collaborating with the Jackson Laboratory to support Alzheimer’s research and working with Professor Baek Min-kyung’s team at Seoul National University on next-generation protein structure prediction.

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

Activision pulls game after PC hacking reports

Activision has removed Call of Duty: WWII from the Microsoft Store and PC Game Pass following reports that hackers exploited a serious vulnerability in the game. Only the PC versions from Microsoft’s platforms are affected, while the game remains accessible via Steam and consoles.

The decision came after several players reported their computers being hijacked during gameplay. Streamed footage showed remote code execution attacks, where malicious code was deployed through the game to seize control of victims’ devices.

AN outdated and insecure build of the game, which had previously been patched elsewhere, was uploaded to the Microsoft platforms. Activision has yet to restore access and continues to investigate the issue.

Call of Duty: WWII was only added to Game Pass in June. The vulnerability highlights the dangers of pushing old game builds without sufficient review, exposing users to significant cybersecurity risks.

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

M&S urges UK firms to report cyberattacks

Marks & Spencer has called for a legal obligation requiring UK companies to report major cyberattacks to national authorities. Chairman Archie Norman told parliament that two serious cyberattacks on prominent firms in recent months had gone unreported.

He argued that underreporting leaves a significant gap in cybersecurity knowledge. It would not be excessive regulation to require companies to report material incidents to the National Cyber Security Centre.

The retailer was hit in April by what is believed to be a ransomware attack involving DragonForce, with links to the Scattered Spider hacking group.

The breach forced a seven-week suspension of online clothing orders, costing the business around £300 million in lost operating profit.

M&S had fortunately doubled its cyber insurance last year, though it may take 18 months to process the claim.

General counsel Nick Folland added that companies must be prepared to operate manually, using pen and paper, when systems go down.

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

Over 2.3 million users hit by Chrome and Edge extension malware

A stealthy browser hijacking campaign has infected over 2.3 million users through Chrome and Edge extensions that appeared safe and even displayed Google’s verified badge.

According to cybersecurity researchers at Koi Security, the campaign, dubbed RedDirection, involves 18 malicious extensions offering legitimate features like emoji keyboards and VPN tools, while secretly tracking users and backdooring their browsers.

One of the most popular extensions — a colour picker developed by ‘Geco’ — continues to be available on the Chrome and Edge stores with thousands of positive reviews.

While it works as intended, the extension also hijacks sessions, records browsing activity, and sends data to a remote server controlled by attackers.

What makes the campaign more insidious is how the malware was delivered. The extensions began as clean, valuable tools, but malicious code was quietly added during later updates.

Due to how Google and Microsoft handle automatic updates, most users receive spyware without taking action or clicking anything.

Koi Security’s Idan Dardikman describes the campaign as one of the largest documented. Users are advised to uninstall any affected extensions, clear browser data, and monitor accounts for unusual activity.

Despite the serious breach, Google and Microsoft have not responded publicly.

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

Grok AI chatbot suspended in Turkey following court order

A Turkish court has issued a nationwide ban on Grok, the AI chatbot developed by Elon Musk’s company xAI, following recent developments involving the platform.

The ruling, delivered on Wednesday by a criminal court in Ankara, instructed Turkey’s telecommunications authority to block access to the chatbot across the country. The decision came after public filings under Turkey’s internet law prompted a judicial review.

Grok, which is integrated into the X platform (formerly Twitter), recently rolled out an update to make the system more open and responsive. The update has sparked broader global discussions about the challenges of moderating AI-generated content in diverse regulatory environments.

In a brief statement, X acknowledged the situation and confirmed that appropriate content moderation measures had been implemented in response. The ban places Turkey among many countries examining the role of generative AI tools and the standards that govern their deployment.

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

AI-powered imposter poses as US Secretary of State Rubio

An imposter posing as US Secretary of State Marco Rubio used an AI-generated voice and text messages to contact high-ranking officials, including foreign ministers, a senator, and a state governor.

The messages, sent through SMS and the encrypted app Signal, triggered an internal warning across the US State Department, according to a classified cable dated 3 July.

The individual created a fake Signal account using the name ‘Marco.Rubio@state.gov’ and began contacting targets in mid-June.

At least two received AI-generated voicemails, while others were encouraged to continue the chat via Signal. US officials said the aim was likely to gain access to sensitive information or compromise official accounts.

The State Department confirmed it is investigating the breach and has urged all embassies and consulates to remain alert. While no direct cyber threat was found, the department warned that shared information could still be exposed if targets were deceived.

A spokesperson declined to provide further details for security reasons.

The incident appears linked to a broader wave of AI-driven disinformation. A second operation, possibly tied to Russian actors, reportedly targeted Gmail accounts of journalists and former officials.

The FBI has warned of rising cases of ‘smishing’ and ‘vishing’ involving AI-generated content.

Experts now warn that deepfakes are becoming harder to detect, as the technology advances faster than defences.

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

The rise and risks of synthetic media

Synthetic media transforms content creation across sectors

The rapid development of AI has enabled significant breakthroughs in synthetic media, opening up new opportunities in healthcare, education, entertainment and many more.

Instead of relying on traditional content creation, companies are now using advanced tools to produce immersive experiences, training simulations and personalised campaigns. But what exactly is synthetic media?

Seattle-based ElastixAI raised $16 million to build a platform that improves how large language models run, focusing on efficient inference rather than training.

Synthetic media refers to content produced partly or entirely by AI, including AI-generated images, music, video and speech. Tools such as ChatGPT, Midjourney and voice synthesisers are now widely used in both creative and commercial settings.

The global market for synthetic media is expanding rapidly. Valued at USD 4.5 billion in 2023, it is projected to reach USD 16.6 billion by 2033, driven mainly by tools that convert text into images, videos or synthetic speech.

The appeal lies in its scalability and flexibility: small teams can now quickly produce a wide range of professional-grade content and easily adapt it for multiple audiences or languages.

However, as synthetic media becomes more widespread, so do the ethical challenges it poses.

How deepfakes threaten trust and security

The same technology has raised serious concerns as deepfakes – highly realistic but fake audio, images and videos – become harder to detect and more frequently misused.

Deepfakes, a subset of synthetic media, go a step further by creating content that intentionally imitates real people in deceptive ways, often for manipulation or fraud.

The technology behind deepfakes involves face swapping through variational autoencoders and voice cloning via synthesised speech patterns. The entry barrier is low, making these tools accessible to the general public.

computer keyboard with red deepfake button key deepfake dangers online

First surfacing on Reddit in 2017, deepfakes have quickly expanded into healthcare, entertainment, and education, yet they also pose a serious threat when misused. For example, a major financial scam recently cost a company USD 25 million due to a deepfaked video call with a fake CFO.

Synthetic media fuels global political narratives

Politicians and supporters have often openly used generative AI to share satirical or exaggerated content, rather than attempting to disguise it as real.

In Indonesia, AI even brought back the likeness of former dictator Suharto to endorse candidates, while in India, meme culture thrived but failed to significantly influence voters’ decisions.

In the USA, figures like Elon Musk and Donald Trump have embraced AI-generated memes and voice parodies to mock opponents or improve their public image.

AI, US elections, Deepfakes

While these tools have made it easier to create misinformation, researchers such as UC Berkeley’s Hany Farid argue that the greater threat lies in the gradual erosion of trust, rather than a single viral deepfake.

It is becoming increasingly difficult for users to distinguish truth from fiction, leading to a contaminated information environment that harms public discourse. Legal concerns, public scrutiny, and the proliferation of ‘cheapfakes’—manipulated media that do not rely on AI—may have limited the worst predictions.

Nonetheless, experts warn that the use of AI in campaigns will continue to become more sophisticated. Without clear regulation and ethical safeguards, future elections may not be able to prevent the disruptive influence of synthetic media as easily.

Children use AI to create harmful deepfakes

School-aged children are increasingly using AI tools to generate explicit deepfake images of their classmates, often targeting girls. What began as a novelty has become a new form of digital sexual abuse.

With just a smartphone and a popular app, teenagers can now create and share highly realistic fake nudes, turning moments of celebration, like a bat mitzvah photo, into weapons of humiliation.

Rather than being treated as simple pranks, these acts have severe psychological consequences for victims and are leaving lawmakers scrambling.

Educators and parents are now calling for urgent action. Instead of just warning teens about criminal consequences, schools are starting to teach digital ethics, consent, and responsible use of technology.

kids using laptops in class

Programmes that explain the harm caused by deepfakes may offer a better path forward than punishment alone. Experts say the core issues—respect, agency, and safety—are not new.

The tools may be more advanced, but the message remains the same: technology must be used responsibly, not to exploit others.

Deepfakes become weapons of modern war

Deepfakes can also be deployed to sow confusion, falsify military orders, and manipulate public opinion. While not all such tactics will succeed, their growing use in psychological and propaganda operations cannot be ignored.

Intelligence agencies are already exploring how to integrate synthetic media into information warfare strategies, despite the risk of backfiring.

A new academic study from University College Cork examined how such videos spread on social media and how users reacted.

While many responded with scepticism and attempts at verification, others began accusing the real footage of being fake. The growing confusion risks creating an online environment where no information feels trustworthy, exactly the outcome hostile actors might seek.

While deception has long been part of warfare, deepfakes challenge the legal boundaries defined by international humanitarian law.

 Crowd, Person, Adult, Male, Man, Press Conference, Head, Face, People

Falsifying surrender orders to launch ambushes could qualify as perfidy—a war crime—while misleading enemies about troop positions may remain lawful.

Yet when civilians are caught in the crossfire of digital lies, violations of the Geneva Conventions become harder to ignore.

Regulation is lagging behind the technology, and without urgent action, deepfakes may become as destructive as conventional weapons, redefining both warfare and the concept of truth.

The good side of deepfake technology

Yet, not all applications are harmful. In medicine, deepfakes can aid therapy or generate synthetic ECG data for research while protecting patient privacy. In education, the technology can recreate historical figures or deliver immersive experiences.

Journalists and human rights activists also use synthetic avatars for anonymity in repressive environments. Meanwhile, in entertainment, deepfakes offer cost-effective ways to recreate actors or build virtual sets.

These examples highlight how the same technology that fuels disinformation can also be harnessed for innovation and the public good.

Governments push for deepfake transparency

However, the risks are rising. Misinformation, fraud, nonconsensual content, and identity theft are all becoming more common.

The danger of copyright infringement and data privacy violations also looms large, particularly when AI-generated material pulls content from social media or copyrighted works without permission.

Policymakers are taking action, but is it enough?

The USA has banned AI robocalls, and Europe’s AI Act aims to regulate synthetic content. Experts emphasise the need for worldwide cooperation, with regulation focusing on consent, accountability, and transparency.

eu artificial intelligence act 415652543

Embedding watermarks and enforcing civil liabilities are among the strategies being considered. To navigate the new landscape, a collaborative effort across governments, industry, and the public is crucial, not just to detect deepfakes but also to define their responsible use.

Some emerging detection methods include certifying content provenance, where creators or custodians attach verifiable information about the origin and authenticity of media.

Automated detection systems analyse inconsistencies in facial movements, speech patterns, or visual blending to identify manipulated media. Additionally, platform moderation based on account reputation and behaviour helps filter suspicious sources.

Systems that process or store personal data must also comply with privacy regulations, ensuring individuals’ rights to correct or erase inaccurate data.

Yet, despite these efforts, many of these systems still struggle to reliably distinguish synthetic content from real one.

As detection methods lag, some organisations like Reality Defender and Witness work to raise awareness and develop countermeasures.

The rise of AI influencers on social media

Another subset of synthetic media is the AI-generated influencers. AI (or synthetic) influencers are virtual personas powered by AI, designed to interact with followers, create content, and promote brands across social media platforms.

Unlike traditional influencers, they are not real people but computer-generated characters that simulate human behaviour and emotional responses. Developers use deep learning, natural language processing, and sophisticated graphic design to make these influencers appear lifelike and relatable.

Finfluencers face legal action over unregulated financial advice.

Once launched, they operate continuously, often in multiple languages and across different time zones, giving brands a global presence without the limitations of human engagement.

These virtual influencers offer several key advantages for brands. They can be precisely controlled to maintain consistent messaging and avoid the unpredictability that can come with human influencers.

Their scalability allows them to reach diverse markets with tailored content, and over time, they may prove more cost-efficient due to their ability to produce content at scale without the ongoing costs of human talent.

Brands can also experiment with creative storytelling in new and visually compelling ways that might be difficult for real-life creators.

Synthetic influencers have also begun appearing in the healthcare sector, although their widespread popularity in the sector remains limited. However, it is expected to grow rapidly.

Their rise also brings significant challenges. AI influencers lack genuine authenticity and emotional depth, which can hinder the formation of meaningful connections with audiences.

Their use raises ethical concerns around transparency, especially if followers are unaware that they are interacting with AI.

Data privacy is another concern, as these systems often rely on collecting and analysing large amounts of user information to function effectively.

Additionally, while they may save money in the long run, creating and maintaining a sophisticated AI influencer involves a substantial upfront investment.

Study warns of backlash from synthetic influencers

A new study from Northeastern University urges caution when using AI-powered influencers, despite their futuristic appeal and rising prominence.

While these digital figures may offer brands a modern edge, they risk inflicting greater harm on consumer trust compared to human influencers when problems arise.

The findings show that consumers are more inclined to hold the brand accountable if a virtual influencer promotes a faulty product or spreads misleading information.

Rather than viewing these AI personas as independent agents, users tend to see them as direct reflections of the company behind them. Instead of blaming the influencer, audiences shift responsibility to the brand itself.

Interestingly, while human influencers are more likely to be held personally liable, virtual influencers still cause deeper reputational damage.

 Accessories, Jewelry

People assume that their actions are fully scripted and approved by the business, making any error seem deliberate or embedded in company practices rather than a personal mistake.

Regardless of the circumstances, AI influencers are reshaping the marketing landscape by providing an innovative and highly adaptable tool for brands. While they are unlikely to replace human influencers entirely, they are expected to play a growing role in digital marketing.

Their continued rise will likely force regulators, brands, and developers to establish clearer ethical standards and guidelines to ensure responsible and transparent use.

Shaping the future of synthetic media

In conclusion, the growing presence of synthetic media invites both excitement and reflection. As researchers, policymakers, and creators grapple with its implications, the challenge lies not in halting progress but in shaping it thoughtfully.

All forms of synthetic media, like any other form of technology, have a dual capacity to empower and exploit, demanding a new digital literacy — one that prioritises critical engagement, ethical responsibility, and cross-sector collaboration.

On the one hand, deepfakes threaten democratic stability, information integrity, and civilian safety, blurring the line between truth and fabrication in conflict, politics, and public discourse.

On the other hand, AI influencers are transforming marketing and entertainment by offering scalable, controllable, and hyper-curated personas that challenge notions of authenticity and human connection.

Rather than fearing the tools themselves, we as human beings need to focus on cultivating the norms and safeguards that determine how, and for whom, they are used. Ultimately, these tools are meant to enhance our way of life, not undermine it.

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