Topic
AI and digital
How fast is AI advancing, and how safe are the systems we depend on?
Each number comes from the dashboard it links to, with the date of the data. Updated every night.
What this is about
Digital risks are reported unevenly. A data breach becomes public when a law requires it or someone finds out. A software flaw enters the US cybersecurity agency’s catalogue once it is known to have been used in attacks. These pages use the registers that exist and say what they miss.
The first page follows that catalogue: the software flaws attackers are exploiting, whose software they are in, and how fast they must be fixed. The second counts the data breaches that became public, from Have I Been Pwned and the notices filed in California. A third follows AI models and the chips they run on, from Epoch AI’s open data: how much computing power the largest models use, who builds them and how many can be downloaded. A fourth counts the incidents of harm from AI recorded in the AI Incident Database.
The dashboards
AI Incidents
How often are AI systems reported to cause harm, what kind of harm, and whose systems? Incidents by year and developer, from the AI Incident Database.
Open dashboard →Daily data · checked 22 Nov 2026AI Models and Chips
How fast is AI growing, who builds the models, and can you download them? Training compute, notable models by country, open weights and AI chips, from Epoch AI.
Open dashboard →Daily data · checked 22 Nov 2026Data Breaches
Are data breaches becoming more common? Breaches added to Have I Been Pwned each year, how many accounts they exposed and what leaked, and the breach notices filed with California's Attorney General since 2012.
Open dashboard →Daily data · checked 22 Nov 2026Exploited Vulnerabilities Monitor
Which software flaws are attackers exploiting right now? Every flaw in CISA's catalogue of known exploited vulnerabilities: how many are added each month, from which vendors, what kind of flaw, and how fast US agencies must fix them.
Open dashboard →Why exploited flaws
Tens of thousands of software flaws are disclosed every year, far more than anyone can patch at once, and only a small share is ever used in real attacks. The exploited vulnerabilities page follows the ones that are, as confirmed by CISA, the US cybersecurity agency.
Most of the flaws CISA adds are new, disclosed the same year, and since June 2026 most new entries give US federal agencies three days to fix them.
Breaches made public
Nobody counts every data breach. The data breaches page uses two open records: the breaches loaded into Have I Been Pwned, the free service where people check their email address, and the notices organisations must file in California when a breach affects more than 500 residents.
Have I Been Pwned added more breaches in 2026 than in any full year before, and California's notices rose too. Recent breaches leak passwords less often than in the 2010s and phone numbers and home addresses more often.
Models and chips
The AI models and chips page uses Epoch AI’s records of notable models. The computing power used to train the largest models has grown several times over each year, and most notable models come from US organisations, with Chinese ones second.
Training compute measures scale, not how capable or safe a model is. Most figures are Epoch’s estimates, because few companies publish them.
When AI causes harm
The AI Incident Database records cases where AI systems caused or nearly caused harm, from news reports and submissions. The AI incidents page counts them by year, kind of harm and the companies named.
Recorded incidents have grown with the use of AI. These are allegations that someone reported; the rise reflects more AI in use and more attention as well as more harm.
How the pages fit together
The electricity and water that AI uses are on the AI energy page under Energy and emissions.
What these pages can’t tell you
Registers of breaches, flaws and incidents show what was reported, not everything that happened. A rising count can mean more attacks or better reporting.
Latest from these dashboards
News, official reports and research picked for each dashboard’s Latest tab and approved by hand.
- Separating recognition from prediction in a Known Exploited Vulnerabilities benchmark: A landmark-corrected Microsoft Windows case study
- Loblaw CEO apologizes to 2 million shoppers for personal data hack, offers free credit monitoring
- Business Associates’ Involvement in US Health Care Data Breaches: Longitudinal Analysis
- AI Incident Database
- Data on AI models
- Have I Been Pwned: all breached websites
- Data security breaches
- Known Exploited Vulnerabilities catalogue
Where the numbers come from
Every source these dashboards use, linked to the publisher. Each dashboard’s Report tab says what it takes from each one.