Network security

Google API keys exposed after Gemini privilege expansion

Security researchers warn that exposed Google API keys in public client-side code could be used to authenticate with the Gemini AI assistant and access private data. The issue arose after developers enabled the Generative Language API in existing projects without updating key permissions.

Truffle Security scanned the November 2025 Common Crawl dataset and identified more than 2,800 live Google API keys publicly exposed in website source code. Some belonged to financial institutions, security firms, recruitment companies, and Google infrastructure.

Before Gemini’s launch, Google Cloud API keys were widely treated as non-sensitive identifiers for services such as Maps, YouTube embeds, analytics, and Firebase. After Gemini was introduced, those duplicate Google API keys also acted as authentication credentials for the AI assistant, expanding their privileges.

Researchers demonstrated the risk by using one exposed key to query the Gemini API models endpoint and list available models. They warned that attackers could exploit such access to extract private data or generate substantial API charges on victim accounts.

Google was notified in November 2025 and later classified the issue as a single-service privilege escalation. The company said it has introduced controls to block leaked keys, limit new AI Studio keys to Gemini-only scope, and notify developers of detected exposure.

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

Financial crime risks are reshaped by the rise of autonomous AI agents

Autonomous AI agents are transforming finance by executing transactions independently and speeding up workflows in digital assets and programmable finance. Software can manage wallets and move funds across blockchains in seconds, narrowing detection windows.

AI agents don’t create new crimes but increase speed and complexity, making accountability essential. Responsibility rests with developers, operators, and beneficiaries, with investigators tracing control, configuration, and economic benefit to determine liability.

Weak oversight or misconfigured rules can lead to significant compliance and enforcement consequences.

Investigations face new challenges as autonomous agents operate across multiple blockchains, decentralised exchanges, and global jurisdictions.

Real-time analytics and automated tracing are essential to link transactions to accountable actors before funds move. Governance architecture and monitoring systems increasingly serve as evidence in regulatory or criminal actions.

Institutions and law enforcement are using AI monitoring, anomaly detection, and automated containment systems. Autonomous AI impacts sanctions and national security, emphasising the need for human oversight alongside automation.

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

ChatGPT Health under fire after study finds major failures in emergency detection

A new evaluation of ChatGPT Health has raised major safety concerns after researchers found it frequently failed to recognise urgent medical emergencies.

The independent study, published in Nature Medicine, reported that the system under-triaged more than half of the clinical scenarios tested, giving advice that could have delayed life-saving treatment.

The research team, led by Ashwin Ramaswamy, created sixty patient simulations ranging from minor illnesses to life-threatening conditions.

Three doctors agreed on the appropriate urgency for each case before comparing their judgement with the model’s responses. The AI performed adequately in straightforward emergencies such as strokes, yet frequently minimised danger in more complex presentations, including severe asthma and diabetic crises.

Experts also warned that ChatGPT Health struggled to detect suicidal ideation reliably. Minor changes to scenario details, such as adding normal lab results, caused safeguards to disappear entirely.

Critics, including health-misinformation researcher Alex Ruani, described the behaviour as dangerously inconsistent and capable of creating a false sense of security.

OpenAI said the study did not reflect typical real-world use but acknowledged the need for continued research and improvement.

Policy specialists argue that the findings underline the need for clear safety standards, external audits and stronger transparency requirements for AI systems operating in sensitive medical contexts.

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

AI misuse in online scams involving OpenAI models

OpenAI has reported new instances of its models being exploited in online scams and coordinated information campaigns. The company detailed actions to remove offending accounts and strengthen safeguards, highlighting misuse in fraud and deceptive content creation.

Several cases involved romance and ‘task’ scams, in which AI-generated messages built emotional engagement before requesting payment. One network, dubbed ‘Operation Date Bait,’ used chatbots to promote a fictitious dating service targeting young men in Indonesia.

Another, ‘Operation False Witness,’ saw actors posing as legal professionals to solicit advance fees for non-existent recovery services.

The report also outlined coordinated campaigns leveraging AI to produce articles, social media posts, and comments on geopolitical topics. In ‘Operation Trolling Stone,’ AI-generated content on a Russian arrest in Argentina was shared widely in multiple languages to mimic grassroots engagement.

OpenAI stressed that AI was sometimes used, but reach and account size largely drove engagement.

The company continues monitoring misuse and collaborates with partners and authorities to curb fraudulent or deceptive activity. Systems have been updated to decline policy-violating requests, and not all suspicious content online was generated using its tools.

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

AT&T data breach settlement wins preliminary approval in $177 million deal

A federal judge in Texas has preliminarily approved a $177 million settlement resolving claims that AT&T failed to safeguard consumer data in two separate breaches. The company denies wrongdoing but agreed to establish compensation funds covering affected customers nationwide.

The agreement creates two non-reversionary funds: $149 million for individuals whose personal data appeared on the dark web, and $28 million for customers whose call and text logs were accessed. It covers a March 2024 breach and a separate incident between May 2022 and early 2023.

Eligible class members may submit claims for cash payments, with amounts depending on the number of valid submissions, and may also receive up to 24 months of credit monitoring. The deadline to opt out or object is 17 October 2025, with a final approval hearing set for 3 December 2025.

Legal and administrative costs, attorneys’ fees, and service awards will be paid from the settlement funds. The case resolves claims brought on behalf of all living US residents whose data was exposed in the two AT&T breaches.

The settlement follows other recent legal challenges facing AT&T, including class actions filed by New York pensioners alleging the company misled investors about the environmental impact of its lead-sheathed cables.

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

Italy orders Amazon to stop processing sensitive employee data after privacy ruling

The Italian data protection authority has ordered Amazon Italia Logistics to halt processing of sensitive employee data after investigators found that the company gathered details ranging from health conditions to union involvement.

Information about workers’ private lives and family members had also been collected, often retained for a decade through internal tracking systems rather than being limited to what labour rules in Italy allow.

Regulators discovered that some data originated from cameras positioned near restrooms and staff break areas, a practice that breached EU privacy standards.

The watchdog concluded that the company’s monitoring went far beyond what employers are permitted to compile when assessing staff performance or workplace needs.

Amazon responded by stressing that protecting employee information remains a priority and said that internal rules and training programmes are designed to ensure compliance. The company added that any findings from the Italian authority would prompt a review of its procedures instead of being dismissed.

An order that arrives as Amazon attempts to regain its lobby badges at the European Parliament.

Access was suspended in 2024 after senior representatives declined to attend hearings on warehouse working conditions, and opposition from MEPs continues to place pressure on Parliament President Roberta Metsola to reject reinstatement.

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

EU moves to enforce digital fairness rules with stronger consumer oversight

Regulatory scrutiny of the EU’s digital fairness framework is set to begin on 1 July as the European Commission moves to tighten its supervision of online platforms.

An initiative that forms part of a broader effort to ensure stronger consumer protection across digital markets, with officials signalling stricter oversight of commercial practices that disadvantage users.

The Commission is preparing a major upgrade of its consumer protection framework, expected by December 2026.

The reforms aim to reinforce enforcement tools under the Unfair Commercial Practices Directive and the Consumer Protection Cooperation Regulation, allowing regulators to intervene more effectively when platforms breach fairness standards.

Michael McGrath, Commissioner for Democracy, Justice and Rule of Law, has highlighted the need for greater transparency and accountability as digital markets expand rapidly.

The forthcoming scrutiny focuses on ensuring that platforms respect transparency obligations, avoid manipulating users and provide fair conditions in online transactions.

Regulators seek to replace fragmented enforcement with a more coordinated model that reflects the increasingly cross-border nature of digital commerce.

Stronger consumer safeguards are becoming central to the digital agenda of the EU.

The next phase of reforms is expected to streamline investigations across member states and deliver more predictable outcomes for affected consumers, offering steadier enforcement instead of reactive measures taken after violations escalate.

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

Microsoft backs Australia’s next phase of digital government with new AI and cloud agreement

Australia’s rise to second place in the OECD Digital Government Index signals renewed momentum for national digital transformation.

A shift that comes as Microsoft signs a new five-year Volume Sourcing Arrangement with the Federal Government, designed to underpin modernisation across public services and create a secure, future-ready foundation for responsible AI adoption.

The agreement led by the Digital Transformation Agency gives agencies access to Microsoft Copilot, Azure, Microsoft 365, Dynamics 365 and a strengthened security and compliance framework instead of continuing reliance on ageing systems.

The arrangement sets clearer strategic pathways for innovation, procurement and skills development through an enhanced governance structure.

It recommits both sides to national security requirements, including the Security of Critical Infrastructure legislation, the Cloud Hosting Certification Framework and IRAP.

These measures allow agencies to expand AI use while retaining control of data and meeting the expectations placed on government institutions.

A successful Copilot trial in 2024 already demonstrated personal productivity gains of around one hour per day for participating staff.

Microsoft is also establishing a $1.55 million training fund for the Australian Public Service to support capability building in ethical AI use and modern cloud operations.

The company emphasises that Australia’s partner ecosystem will gain new opportunities because the agreement simplifies how local firms engage with government agencies. Such an approach forms an important part of the wider public sector reform agenda announced last year.

The new deal aligns with national priorities set out in the Whole-of-Government Cloud Computing Policy and the National AI Plan.

Australia now enters a pivotal period in which digital transformation is guided not only by technological capacity but by the frameworks of trust, resilience and public benefit that shape how government services evolve.

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

AI misuse exposed as OpenAI details global disinformation and scam networks

OpenAI said criminal and state-linked groups misused ChatGPT for disinformation, scams and covert influence. Its latest threat report details coordinated account bans and highlights how AI tools are embedded within broader operational workflows rather than used in isolation.

One investigation linked accounts to Chinese law enforcement engaged in what were described as ‘cyber special operations’. Activities included planning influence campaigns, mass-reporting dissidents and drafting forged materials, with related efforts continuing through other tools despite model refusals.

The report also outlined a Cambodia-based romance scam targeting young men in Indonesia through a fake dating agency. Operators combined manual prompting with automated chatbots to sustain conversations and facilitate financial fraud, leading to account removals.

Separately, accounts tied to Russia’s ‘Rybar’ network used ChatGPT to draft and translate posts distributed across multiple platforms. OpenAI noted that campaign impact depended more on account reach and coordination than on AI-generated content alone.

Across China, Russia and parts of Southeast Asia, actors treated AI as one tool among many, alongside fake profiles, paid advertising and forged documents. OpenAI called for cross-industry vigilance, stressing the need to analyse behavioural patterns across platforms.

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

Meta AI flood of unusable abuse tips overwhelms US investigators

Investigators in the US say that AI used by Meta is flooding child protection units with large volumes of unhelpful reports, thereby draining resources rather than assisting ongoing cases.

Officers in the Internet Crimes Against Children network told a New Mexico court that most alerts generated by the company’s platforms lack essential evidence or contain material that is not criminal, leaving teams unable to progress investigations.

Meta rejects the claim that it prioritises profit, stressing its cooperation with law enforcement and highlighting rapid response times to emergency requests.

Its position is challenged by officers who say the volume of AI-generated alerts has doubled since 2024, particularly after the Report Act broadened reporting obligations.

They argue that adolescent conversations and incomplete data now form a sizeable portion of the alerts, while genuine cases of child sexual abuse material are becoming harder to detect.

Internal company documents disclosed at trial show Meta executives raising concerns as early as 2019 about the impact of end-to-end encryption on the firm’s ability to identify child exploitation and support investigators.

Child safety groups have long warned that encryption could limit early detection, even though Meta says it has introduced new tools designed to operate safely within encrypted environments.

The growing influx of unusable tips is taking a heavy toll on investigative teams. Officers in the US say each report must still be reviewed manually, despite the low likelihood of actionable evidence, and this backlog is diminishing morale at a time when they say resources have not kept pace with demand.

They warn that meaningful cases risk being delayed as units struggle with a workload swollen by AI systems tuned to avoid regulatory penalties rather than investigative value.

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

Diplo chatbot can make mistakes. Consider checking important information.