OpenAI has rolled out an update to ChatGPT Voice that unifies voice and text in a single interface. Users can now speak, type or mix both without switching screens mid-conversation.
The redesigned chat window displays live transcriptions and responses in real-time. Users can scroll through earlier messages and view images, maps and other visuals while the exchange continues in one place.
You can now use ChatGPT Voice right inside chat—no separate mode needed.
You can talk, watch answers appear, review earlier messages, and see visuals like images or maps in real time.
Previously, voice required a separate mode that hid the main chat history and shared content. OpenAI says the unified layout should make longer, mixed-mode conversations feel more natural and less fragmented.
Voice and text can still be used interchangeably, but ending a voice session requires tapping ‘End’ before returning to text-only use. Those who prefer the old layout can re-enable a separate voice view in settings.
The revamped Voice experience is becoming the default on web and mobile apps as the update rolls out. OpenAI aims to make ChatGPT feel more like a flexible conversational assistant across various devices.
Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot!
OpenAI says a security incident at Mixpanel exposed limited metadata linked to the API interface. Mixpanel’s systems, not OpenAI’s, were compromised during the intrusion. No chat content, passwords, API keys, or payment information was affected.
Mixpanel told OpenAI that an attacker exported a dataset containing basic user profile fields. The information includes names, email addresses, coarse location data, and browser details. OpenAI has removed Mixpanel from production and is notifying impacted users.
OpenAI maintains that its internal infrastructure remains secure with no evidence of unauthorised access. Wider reviews across the vendor ecosystem are underway to assess potential risks. The company has raised security requirements for partners and continues to monitor for misuse.
Security teams warn that the leaked data could fuel phishing or social-engineering attempts. Users are urged to treat unsolicited messages with caution and verify communications sent under the OpenAI name. Multi-factor authentication remains strongly recommended for all accounts as an added safeguard.
OpenAI reiterates that trust and privacy remain core to its products and operations. The organisation has ended its use of Mixpanel and is reviewing supporting services to prevent similar issues. Impacted organisations will receive direct notifications as the investigation continues.
Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot!
The EU member states have endorsed a position for new rules to counter child sexual abuse online. The plan introduces duties for digital services to prevent the spread of abusive material. It also creates an EU Centre to coordinate enforcement and support national authorities.
Service providers must assess how their platforms could be misused and apply mitigation measures. These may include reporting tools, stronger privacy defaults for minors, and controls over shared content. National authorities will review these steps and can order additional action where needed.
A three-tier risk system will categorise services as high, medium, or low risk. High-risk platforms may be required to help develop protective technologies. Providers that fail to comply with obligations could face financial penalties under the regulation.
Victims will be able to request the removal or disabling of abusive material depicting them. The EU Centre will verify provider responses and maintain a database to manage reports. It will also share relevant information with Europol and law enforcement bodies.
The Council supports extending voluntary scanning for abusive content beyond its current expiry. Negotiations with the European Parliament will now begin on the final text. The Parliament adopted its position in 2023 and will help decide the Centre’s location.
Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot!
UNESCO is strengthening capacities in AI ethics and regulation across Ecuador and Latin America through two newly launched courses. The initiatives aim to enhance digital governance and ensure the ethical use of AI in the region.
The first course, ‘Regulation of Artificial Intelligence: A View from and towards Latin America,’ is taking place virtually from 19 to 28 November 2025.
Organised by UNESCO’s Social and Human Sciences Sector in coordination with UNESCO-Chile and CTS Lab at FLACSO Ecuador, the programme involves 30 senior officials from key institutions, including the Ombudsman’s Office and the Superintendency for Personal Data Protection.
Participants are trained on AI ethical principles, risks, and opportunities, guided by UNESCO’s 2021 Recommendation on the Ethics of AI.
The ‘Ethical Use of AI’ course starts next week for telecom and electoral officials. The 20-hour hybrid programme teaches officials to use UNESCO’s RAM to assess readiness and plan ethical AI strategies.
UNESCO aims to train 60 officials and strengthen AI ethics and regulatory frameworks in Ecuador and Chile. The programmes reflect a broader commitment to building inclusive, human-rights-oriented digital governance in Latin America.
Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot!
Quantum computing has long been framed as a future promise, but D-Wave argues real-world use has now arrived. The company says its Advantage2 system is already running complex optimisation tasks for businesses through both cloud and on-premise deployment.
D-Wave highlights a recent physics experiment as evidence of this shift, claiming the system solved a materials-modelling problem that would take a top supercomputer nearly a million years. The result, completed in minutes, serves as a proof point of practical quantum performance.
The company says accessibility is central to its approach, emphasising that Advantage2 can be programmed in Python without specialist quantum expertise. It frames this ease of use as essential to broader adoption beyond research labs.
Industry deployments are cited across logistics, telecoms, and manufacturing. D-Wave points to scheduling gains at Pattison Food Group, network optimisation at NTT Docomo, and faster production planning at Ford Otosan as examples of measurable operational benefits.
Energy efficiency is another focus, with D-Wave stating that each of its six hardware generations draws roughly 12.5 kilowatts. The company argues that this stable power use, paired with rising performance, positions quantum systems as a lower-energy option for hard computational problems.
Would you like to learn more about AI, tech, and digital diplomacy? If so, ask our Diplo chatbot!
Multiple London councils are responding to a cyberattack that has disrupted shared IT systems and raised concerns about data exposure. Kensington and Chelsea and Westminster councils detected the incident on Monday and alerted the Information Commissioner’s Office as investigations began.
The councils say they are working with specialist incident teams and the National Cyber Security Centre (NCSC) to protect systems and keep key services running. Several platforms have been affected, and staff have been redeployed to support residents through monitored phone lines and email channels.
Hammersmith and Fulham, which shares IT services with the affected councils, has also reported disruption. Local leaders say it is too early to confirm who was responsible or whether personal data has been compromised. Overnight mitigation work has been carried out as monitoring continues.
Security researchers describe indications of a serious intrusion involving lateral movement across shared infrastructure. They warn that attackers may escalate to data theft or encryption, given the sensitivity of the information held by local authorities.
National security agencies and police are assessing the incident’s potential impact. Analysts say the attack highlights long-standing risks facing councils that manage extensive services on limited budgets and with inconsistent cyber safeguards.
Would you like to learn more about AI, tech, and digital diplomacy? If so, ask our Diplo chatbot!
The internet has become part of almost everything we do. It helps us work, stay in touch with friends and family, buy things, plan trips, and handle tasks that would have felt impossible until recently. Most people cannot imagine getting through the day without it.
But there is a hidden cost to all this convenience. Most of the time, online services run smoothly, with countless systems working together in the background. But every now and then, though, a key cog slips out of place.
When that happens, the effects can spread fast, taking down apps, websites, and even entire industries within minutes. These moments remind us how much we rely on digital services, and how quickly everything can unravel when something goes wrong. It raises an uncomfortable question. Is digital dependence worth the convenience, or are we building a house of cards that could collapse, pulling us back into reality?
Warning shots of the dot-com Era and the infancy of Cloud services
In its early years, the internet saw several major malfunctions that disrupted key online services. Incidents like the Morris worm in 1988, which crashed about 10 percent of all internet-connected systems, and the 1996 AOL outage that left six million users offline, revealed how unprepared the early infrastructure was for growing digital demand.
A decade later, the weaknesses were still clear. In 2007, Skype, then with over 270 million users, went down for nearly two days after a surge in logins triggered by a Windows update overwhelmed its network. Since video calls were still in their early days, the impact was not as severe, and most users simply waited it out, postponing chats with friends and family until the issue was fixed.
As the dot-com era faded and the 2010s began, the shift to cloud computing introduced a new kind of fragility. When Amazon’s EC2 and EBS systems in the US-East region went down in 2011, the outage took down services like Reddit, Quora, and IMDb for days, exposing how quickly failures in shared infrastructure can cascade.
A year later, GoDaddy’s DNS failure took millions of websites offline, while large-scale Gmail disruptions affected users around the world, early signs that the cloud’s growing influence came with increasingly high stakes.
By the mid-2010s, it was clear that the internet had evolved from a patchwork of standalone services to a heavily interconnected ecosystem. When cloud or DNS providers stumbled, their failures rippled simultaneously across countless platforms. The move to centralised infrastructure made development faster and more accessible, but it also marked the beginning of an era where a single glitch could shake the entire web.
Centralised infrastructure and the age of cascading failures
The late 2000s and early 2010s saw a rapid rise in internet use, with nearly 2 billion people worldwide online. As access grew, more businesses moved into the digital space, offering e-commerce, social platforms, and new forms of online entertainment to a quickly expanding audience.
With so much activity shifting online, the foundation beneath these services became increasingly important, and increasingly centralised, setting the stage for outages that could ripple far beyond a single website or app.
The next major hit came in 2016, when a massive DDoS attack crippled major websites across the USA and Europe. Platforms like Netflix, Reddit, Twitter, and CNN were suddenly unreachable, not because they were directly targeted, but because Dyn, a major DNS provider, had been overwhelmed.
The attack used the Mirai botnet malware to hijack hundreds of thousands of insecure IoT devices and flood Dyn’s servers with traffic. It was one of the clearest demonstrations yet that knocking out a single infrastructure provider could take down major parts of the internet in one stroke.
In 2017, another major outage occurred, with Amazon at the centre once again. On 28 February, the company’s Simple Storage Service (S3) went down for about 4 hours, disrupting access across a large part of the US-EAST-1 region. While investigating a slowdown in the billing system, an Amazon engineer accidentally entered a typo in a command, taking more servers offline than intended.
That small error was enough to knock out services like Slack, Quora, Coursera, Expedia and countless other websites that relied on S3 for storage or media delivery. The financial impact was substantial; S&P 500 companies alone were estimated to have lost roughly 150 million dollars during the outage.
Amazon quickly published a clear explanation and apology, but transparency could not undo the economic damage nor (yet another) sudden reminder that a single mistake in a centralised system could ripple across the entire web.
Outages in the roaring 2020s
The S3 incident made one thing clear. Outages were no longer just about a single platform going dark. As more services leaned on shared infrastructure, even small missteps could take down enormous parts of the internet. And this fragility did not stop at cloud storage.
Over the next few years, attention shifted to another layer of the online ecosystem: content delivery networks and edge providers that most people had never heard of but that nearly every website depended on.
The 2020s opened with one of the most memorable outages to date. On 4 October 2021, Facebook and its sister platforms, Instagram, WhatsApp, and Messenger, vanished from the internet for nearly 7 hours after a faulty BGP configuration effectively removed the company’s services from the global routing table.
Millions of users flocked to other platforms to vent their frustration, overwhelming Twitter, Telegram, Discord, and Signal’s servers and causing performance issues across the board. It was a rare moment when a single company’s outage sent measurable shockwaves across the entire social media ecosystem.
But what happens when outages hit industries far more essential than social media? In 2023, the Federal Aviation Administration was forced to delay more than 10,000 flights, the first nationwide grounding of air traffic since the aftermath of September 11.
A corrupted database file brought the agency’s Notice to Air Missions (NOTAM) system to a standstill, leaving pilots without critical safety updates and forcing the entire aviation network to pause. The incident sent airline stocks dipping and dealt another blow to public confidence, showing just how disruptive a single technical failure can be when it strikes at the heart of critical infrastructure.
Outages that defined 2025
The year 2025 saw an unprecedented wave of outages, with server overloads, software glitches and coding errors disrupting services across the globe. The Microsoft 365 suite outage in January, the Southwest Airlines and FAA synchronisation failure in April, and the Meta messaging blackout in July all stood out for their scale and impact.
But the most disruptive failures were still to come. In October, Amazon Web Services suffered a major outage in its US-East-1 region, knocking out everything from social apps to banking services and reminding the world that a fault in a single cloud region can ripple across thousands of platforms.
Just weeks later, the Cloudflare November outage became the defining digital breakdown of the year. A logic bug inside its bot management system triggered a cascading collapse that took down social networks, AI tools, gaming platforms, transit systems and countless everyday websites in minutes. It was the clearest sign yet that when core infrastructure falters, the impact is immediate, global and largely unavoidable.
And yet, we continue to place more weight on these shared foundations, trusting they will hold because they usually do. Every outage, whether caused by a typo, a corrupted file, or a misconfigured update, exposes how quickly things can fall apart when one key piece gives way.
Going forward, resilience needs to matter as much as innovation. That means reducing single points of failure, improving transparency, and designing systems that can fail without dragging everything down. The more clearly we see the fragility of the digital ecosystem, the better equipped we are to strengthen it.
Outages will keep happening, and no amount of engineering can promise perfect uptime. But acknowledging the cracks is the first step toward reinforcing what we’ve built — and making sure the next slipped cog does not bring the whole machine to a stop.
The smoke and mirrors of the digital infrastructure
The internet is far from destined to collapse, but resilience can no longer be an afterthought. Redundancy, decentralisation and smarter oversight need to be part of the discussion, not just for engineers, but for policymakers as well.
Outages do not just interrupt our routines. They reveal the systems we have quietly built our lives around. Each failure shows how deeply intertwined our digital world has become, and how fast everything can stop when a single piece gives way.
Will we learn enough from each one to build a digital ecosystem that can absorb the next shock instead of amplifying it? Only time will tell.
Would you like to learn more about AI, tech, and digital diplomacy? If so, ask our Diplo chatbot!
OpenAI has unveiled an updated ChatGPT interface that combines voice and text features in a single view. Users can speak naturally at any point in a chat and receive responses in text, audio, or images. The new layout also introduces real-time map displays.
The redesign adds a scrolling transcript within the chat window. It allows users to revisit earlier exchanges and move easily between reading and listening. OpenAI states that the goal is to support voice-led tasks without compromising clarity.
With the unified experience, conversations no longer require switching modes. ChatGPT can deliver audio, written, and visual replies simultaneously. Maps and images appear directly alongside the voice response.
Every spoken message is automatically transcribed. However, this helps users follow more extended discussions and keep a record for later reference. OpenAI says the feature supports both accessibility and everyday convenience.
The update is rolling out gradually across web and mobile platforms. Users who prefer the earlier voice-only layout can revert to it in settings. OpenAI says the unified mode will remain the default as development continues.
Would you like to learn more about AI, tech, and digital diplomacy? If so, ask our Diplo chatbot!
The UK’s Met Office has laid out a strategic plan for integrating AI, specifically machine learning (ML), with traditional physics-based climate and weather models. The aim is to deliver what it calls an ‘optimal blend’ of AI-driven and physics-based forecasting.
To clarify what that blend might look like, the Met Office has defined five distinct approaches. One is the familiar independent physics-based model, which uses physical laws to simulate atmospheric dynamics, trusted but computationally intensive.
At the other end is an independent ML-based model that learns patterns entirely from data, offering far greater speed and scalability.
Between these extremes lie two ‘hybrid’ approaches: hybrid-integrated ML, where ML replaces or enhances parts of the physics model, and hybrid-composite ML, where ML and physics models run separately and feed into each other.
A fifth option is augmented ML, where ML is applied after the model has run to improve its output (for example, downscaling or refining ensemble forecasts).
However, this framework is more than a technical taxonomy; it provides a shared language for scientists, policymakers, and clients to understand how AI and traditional modelling can coexist.
It also helps guide future decisions, for example, allowing gradual adoption of ML in places where it makes sense, while preserving the robustness of well-understood physics methods in critical areas.
The move comes as ML-based weather and climate tools have shown increasing promise. For instance, in 2025, the Met Office published research showing a purely ML-based model achieved seasonal forecasting skill comparable to conventional physics-based methods, but with far lower computing demands.
For digital-policy watchers and climate analysts alike, this signals a shift: forecasting may become more dynamic, scalable and accessible, especially valuable in a changing climate where speed, resolution and adaptability matter as much as theoretical accuracy.
Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot!
Amazon Web Services plans to invest $50 billion in high performance AI infrastructure dedicated to US federal agencies. The programme aims to broaden access to AWS tools such as SageMaker AI, Bedrock and model customisation services, alongside support for Anthropic’s Claude.
The expansion will add around 1.3 gigawatts of compute capacity, enabling agencies to run larger models and speed up complex workloads. AWS expects construction of the new data centres to begin in 2026, marking one of its most ambitious government-focused buildouts to date.
Chief executive Matt Garman argues the upgrade will remove long-standing technology barriers within government. The company says enhanced AI capabilities could accelerate work in areas ranging from cybersecurity to medical research while strengthening national leadership in advanced computing.
AWS has spent more than a decade developing secure environments for classified and sensitive government operations. Competitors have also stepped up US public sector offerings, with OpenAI, Anthropic and Google all rolling out heavily discounted AI products for federal use over the past year.
Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot!
