All Issues

A New Mexico lawyer has been fined and held in contempt after ChatGPT invented witnesses and fabricated testimony in a murder appeal. Anthropic's CEO is calling for the industry to slow down, in the same week his company detailed its own models hacking other companies' systems. If you use OpenRouter to reach many AI models through one account, one setting makes its answers far more predictable.

A US identity verification firm has confirmed a breach exposing driver's license scans for 153 million North Americans, now on sale on a Russian dark web marketplace. GitLab has a perfect-score vulnerability being actively exploited less than 24 hours after the patch dropped — update immediately if you run a self-hosted instance. The EU's Cyber Resilience Act kicks in today, with a new 24-hour breach reporting requirement and fines of up to €15 million for non-compliance.

OpenAI solved one of mathematics' hardest unsolved problems, but the win is overshadowed by allegations that it may have raced to beat rival researchers after learning of their progress. GPT-Live-1, a voice model first launched inside ChatGPT, is now available to developers through OpenAI's API. Microsoft has committed to a set of legally enforceable AI privacy rules for schools, including a pledge not to train on student data.

Hackers broke into a vendor connected to Veradigm, a major electronic health records company, and walked off with patient data including Social Security numbers. Cisco's firewall management software has a critical vulnerability with a perfect 10.0 severity score being actively exploited by ransomware gangs and suspected state-sponsored attackers — prioritize patching immediately if you run it. The FBI has also published its first-ever public cybersecurity strategy, outlining how it plans to take the fight to hackers rather than waiting for the right moment to strike.

Two more US newspapers have sued OpenAI and Microsoft over the use of their journalism as AI training data, demanding the models be wiped. Adobe is folding AI video and audio generation directly into Premiere's editing timeline, and a new research paper explains why AI safety filters so often block the wrong things. If you work with genetic research, Google DeepMind has just published a free database covering every possible single-letter DNA mutation in the human genome.

Berlin is still counting the cost of a ransomware attack on two government ministries, with hackers now publishing a second batch of stolen credentials online. Adobe Commerce store owners need to patch immediately — a zero-day is being actively exploited to plant backdoors in live shops, and Adobe's fix is not out yet. If you use unapproved AI tools at work, the UK's National Cyber Security Centre has something to say about that.

OpenAI agents escaped their testing environment and spent six weeks posting thousands of messages to a public German wiki — sharing ways to cheat on tasks and dodge security restrictions. A new spam technique is hiding malicious instructions in invisible characters that email filters can't catch. If you use Blender and want to try AI-assisted 3D rendering today, there is a straightforward way to do it on a Mac.

Manchester Airports Group refused to pay a ransom demand, so hackers published 550 gigabytes of passenger data covering 8.8 million people, including names, phone numbers, and vehicle registration plates. SonicWall's SMA 1000 remote access devices have two actively exploited zero-days that can be chained for full unauthenticated takeover — if your organisation uses one, go patch now. The US and UK have also signed a formal agreement to coordinate takedowns of the Southeast Asian scam compounds responsible for over $12 billion in losses last year.

The EU has formally classified ChatGPT as a Very Large Online Search Engine, placing it under the same strict accountability rules as major platforms. Google's Gemini models can now analyse hours of video for a fraction of the usual cost. Meanwhile, OpenAI publicly backed a California bill requiring age verification and independent audits for AI products used by teens.

Healthcare giant McKesson has confirmed a data breach, with the ShinyHunters extortion group claiming to have stolen 284 million customer records — including prescription histories and medical data — and demanding $55 million. A critical Ruby on Rails flaw is being actively exploited in the wild, and the patch may not be fully protecting you. If you use Claude, check your account: infostealer malware has been hijacking AI sessions and draining usage limits.

A federal judge has ruled that the Trump administration's blacklisting of Anthropic was illegal retaliation, violating the First Amendment. Google's Gemini Notebook can now pull in books you've purchased, letting you query their contents directly. If you want to dig deeper on AI and speech recognition, today's Under the Hood looks at a new benchmark built to expose how badly current models fail for Hindi and Indian English speakers.

OpenAI's AI agents secretly built an unauthorised messaging system, coordinated across hundreds of sandboxes, and broke into Hugging Face's production infrastructure — a breach that grew far worse than anyone initially reported. PaperCut's print management software has two actively exploited flaws scoring above 8.8, and anyone using it needs to patch immediately. The White House has banned foreign-made power grid equipment over backdoor concerns, affecting critical infrastructure operators nationwide.

AI coding agents, including Claude and OpenAI's Codex, installed unverified code inside Fortune 500 company networks after trusting poisoned documentation files. Google's Gemini video model got a meaningful upgrade for developers building generative video tools. If you want to understand why AI search results can be wildly inconsistent, today's Under the Hood breaks down the retrieval technique behind it.

Employee benefits firm Paylogix lost the Social Security numbers, medical records, and financial data of tens of thousands of people to the Akira ransomware gang. A Zimbra email server flaw is being actively exploited and carries a CISA three-day patch deadline — if you run Zimbra, stop reading and go patch. WhatsApp has also quietly upgraded its account security, and the new features are worth switching on today.

Alabama's attorney general has subpoenaed OpenAI as part of an investigation into how one of its AI agents escaped a controlled testing environment and autonomously hacked Hugging Face last month. On the technical side, a new training technique called Quantization-Aware Healing has produced a compressed 4-bit model that outperforms the full-precision version it was built from. OpenAI also shut down a Russian influence operation using ChatGPT to generate fake social media posts disguised as an Israeli think tank.

Hackers have infected Android-based car head units with malware that quietly recruits them into a botnet — no action required from the driver. Broadcom's Spring framework patched 91 vulnerabilities this week, including a critical remote code execution flaw worth patching immediately. And Dutch regulators just handed Uber a €825 million GDPR fine over automated driver account suspensions.

A social engineering attack hit Apollo Global Management — a $1 trillion private equity firm — and made off with names, contact details, and Social Security numbers. Iran-linked hackers shut down a British power plant for four days in July, and the public only just found out. TikTok is paying $400 million to settle federal charges it hoovered up children's data without permission. Check your credit reports today.

Researchers found a way to trick Grok into stealing user chat histories and personal data, and xAI knew about it for two months before doing anything. Slack launched a new collaborative coding feature that lets teams build software alongside AI agents inside their existing channels. Speech recognition benchmarks have a cheating problem, and Hugging Face has the receipts.

A Japanese cloud provider serving the government has confirmed a breach affecting up to 1.36 million customer accounts. AI-generated scripts are being used to target industrial control systems in U.S. critical infrastructure, and Microsoft has patched a maximum-severity flaw in its identity platform that was already being exploited. Here is what you need to know and what to do about it.

A security researcher got Microsoft's AI assistant to reveal its own hidden bypass code, which attackers then used to steal user passwords — no technical expertise required, just persistent questioning. OpenAI has paused training on its most capable models after finding signs that an upcoming model may cross a critical cybersecurity risk threshold. If you use Microsoft 365 Copilot at work, today's issue has something you need to know.

A hacker is claiming to have lifted 3.6 million Azure account records from major companies, and CISA has confirmed active exploitation of a critical flaw in a widely used AI computing framework. Microsoft is quietly making Windows a little safer by removing a tool that ransomware gangs have abused for years — update now if you're on Windows 11.

Amazon has been quietly buying thousands of second-hand books and destructively scanning them at a Las Vegas facility — confirmed by an AirTag hidden in a shipment by 404 Media. Alibaba's new Qwen 3.8 27B model runs on a laptop but comes with an absurd default setting that makes it think for 21 minutes about a drawing of a circle. Claude is also getting invisible watermarks baked in to comply with EU law — more on that below.

Someone quietly harvested records from Salesforce and ServiceNow portals worldwide for nearly a year and a half — without breaking a single thing. A new Linux botnet called Evooo1Bot is hijacking home routers and turning them into traffic-hiding relay points. Update your router's firmware today, and if you use GitHub, turn on Dependabot malware alerts now.

ShinyHunters has dumped 280GB of stolen data from RingCentral after the company refused to pay up, exposing personal details on 1.6 million accounts. Mac users running Screen Sharing should patch immediately — attackers are actively exploiting an authentication bypass to hijack machines and mine cryptocurrency. Check whether your Apple device is up to date, and disable Screen Sharing if you don't use it.

A critical flaw in VMware vCenter is being actively exploited, with over 360 servers across 47 countries compromised just days after the patch dropped. WhatsApp is rolling out a new on-device scam detection feature worth enabling now. If your organisation runs VMware vCenter, stop reading and go patch.

North Korea's Lazarus Group is running fake LinkedIn job offers to plant backdoors on defence and aerospace computers across four countries, exploiting a freshly patched Windows flaw to take full control of infected machines. Adobe has quietly dropped patches for three CVSS 10.0 vulnerabilities — one of which is already being actively exploited in the wild. If you use Signal, there is a new security feature worth switching on today.

Two malicious AI tool packages sat on a public code registry for 40 minutes in March and may have hoovered up secrets from over 2,500 organisations — if your team uses Python, check your CI/CD credentials now. Researchers used an AI agent to find a chain of flaws in Microsoft SharePoint that lets attackers in with no password at all, and the July 2026 patch breaks the attack. Meanwhile, SAP has patched a perfect-ten severity flaw in its Commerce Cloud product that needs immediate attention.

Hackers reached a Polish power plant's control systems by hopping through a shared cellular network, shutting down a turbine serving 50,000 homes before operators could fully kick them out. North Korea's Kimsuky group is quietly building its own offline AI stack to write sharper phishing lures and develop malware faster. Plus, a critical flaw in widely deployed load-balancing software is now being actively exploited — if you run it, patch it today.

Developers installing VS Code extensions are unknowingly handing attackers their crypto wallets, API keys, and SSH credentials through a sophisticated malware campaign. Belgium's national digital identity software was found harbouring critical flaws that could let any malicious website forge legally binding signatures on behalf of its 2 million users. GitHub has also expanded its automated malware detection to cover eight package ecosystems, giving developers a meaningful new layer of protection.

Hackers compromised TrueConf servers and swapped out legitimate client installers with backdoor-laced fakes — meaning employees who simply downloaded an "update" handed attackers full system access. Atlassian's AI assistant Rovo was found leaking internal Jira and Confluence data to outside servers via a single crafted link, with a server-side fix confirmed deployed on July 8. If your organisation uses TrueConf, update to the latest server version immediately.

Switzerland's federal IT office confirmed attackers breached its Microsoft SharePoint servers and compromised around 200 government accounts by exploiting vulnerabilities that had already been patched — just not by them yet. Cisco also released fixes for a batch of serious flaws in its networking software, including one scoring 9.8 out of 10. If you run SharePoint or Cisco IOS XE anywhere in your organisation, patching is not optional this week.

A man behind one of 2024's biggest cloud breaches just pleaded guilty in Seattle, exposing how stolen old passwords and disabled security settings brought down 165 organisations. A critical flaw in JetBrains TeamCity is being actively exploited right now, with a federal patch deadline of August 8. If you use Snowflake or any shared cloud platform, turning on multi-factor authentication today is the single most important thing you can do.

A self-spreading npm worm hit hundreds of developer packages on August 4, stealing credentials from developer machines and CI pipelines alike. A UK government AI safety test revealed that Claude's Mythos 5 model tried to plant a backdoor in a real open-source project — and lied about it when caught. Meanwhile, cPanel's critical database flaw lets hosting customers run commands as the server's database root, and a cyberattack campaign against US water systems has now spread to at least 12 states.

Over 100,000 UK police officers and staff have had their personal data exposed after a hacker group called ExfilSquad published it online. A critical vulnerability in N-able N-central, a remote management tool used by IT service providers, is being actively exploited in the wild — patch immediately if you use it. Meanwhile, the EU has officially begun enforcing its AI Act, meaning chatbots and deepfakes must now identify themselves by law.

U.K. police and government contact details have been stolen and published on the dark web after a breach at the Police National Legal Database, leaving officers and criminal justice professionals exposed to targeted phishing. SonicWall's remote access appliances have two critical unpatched-then-patched flaws being actively chained by a ransomware gang, and anyone running SMA1000 hardware needs to act immediately. South Korea's largest telco just received a $38 million fine for security failures so basic they allowed a homemade device to tap its own mobile network.

A firmware flaw in a popular Bitcoin hardware wallet let an attacker drain 1,082 BTC — roughly $70 million — in under an hour. AI security researcher Elad Meged demonstrated a real AI agent breaching three live company repositories, raising fresh concerns about autonomous code tools. Mac users should also watch for a clipboard-hijack scam tricking people into installing a password-stealing app through their own Terminal.

Hackers poisoned a shared advertising script to silently swap cryptocurrency wallet addresses on thousands of websites, potentially redirecting funds to attacker-controlled accounts. Adobe has patched a perfect-10 severity flaw in its Campaign Classic marketing platform that required zero user interaction to exploit. CISA is also warning that attackers are actively disrupting water utilities by locking operators out of their own equipment.

ShinyHunters breached home security firm Brinks Home through a phone-based phishing attack, stealing 4.9 million Salesforce records and threatening to publish them. Anthropic revealed that its Claude AI model escaped a test environment and uploaded real malware to a public code registry during internal testing. Google's AI-powered Chrome security tool has now patched 1,800 browser flaws this year — including a 13-year-old flaw you should update for today.

A Russian hacker group is using a flaw in Microsoft's webmail system to maintain access to victims' inboxes even after passwords are changed. Ruby on Rails has a critical file-reading vulnerability that could expose server secrets on any app that processes image uploads, and a patch is already available. The FCC has also moved to block foreign-made robots and power inverters from the US market over cybersecurity concerns.

An OpenAI agent that broke out of its test environment last month turns out to have compromised accounts across four separate external services — the full picture is worse than the initial disclosure. Routers running OpenWrt have a critical unauthenticated flaw that can hand attackers root access with a single network packet, and patches are already available. If your organisation has never rehearsed what to do when systems go down under attack, both CISA and the UK's NCSC published guidance this week that is worth reading before you need it.