Nine days before the presidential runoff, Colombia’s digital campaign already has a formal diagnosis: the Electoral Observation Mission (MOE) presented three warnings about the online debate ahead of the June 21 runoff. This is no longer a general concern about deepfakes: it is a report with data, which sorts into three fronts what we had been following as a single diffuse threat.
The first is the escalation of hostility. After the May 31 first round, the monitoring —carried out together with UNDP and the Foundation for Press Freedom— found that the conversation moved from insults to “explicit references to physical violence, direct threats and calls to silence or eliminate” political opponents. The second is disinformation with artificial intelligence: pieces fabricated and distributed with AI, added to rumors of fraud without evidence, which erode fact-based deliberation. The third is the newest in the story: untraceable money. As of May 31, fourteen campaigns reported more than 99.5 billion pesos in income and 85.1 billion in spending, but the MOE was unable to reconcile much of the digital advertising with the declared amounts. In other words: political ads circulating without anyone being able to know clearly who pays for them. Three different problems, a single terrain —social media— and a single deadline: June 21, when Abelardo de la Espriella and Iván Cepeda face each other at the polls.
Also today
- June brings an unprecedented wave of model launches — Three frontier models are aiming for the same month: Gemini 3.5 Pro (announced by Google in May, with general availability promised for June), Claude Sonnet 4.8 and GPT-5.6. Never before had so many top-tier releases lined up in a single 30-day window. In the background, the assistant market share is shifting: ChatGPT is still the leader but fell to 54.7% from 76.5% in early 2025, Gemini climbed to 27.4% and Claude is growing fast from a smaller base.
- Japan’s megabanks add Claude Mythos to their cyber defense — The country’s three largest banks (MUFG, SMBC and Mizuho) are getting access to Mythos, Anthropic’s model capable of detecting and simulating zero-day vulnerabilities. It is the first access by Japanese entities to this frontier cybersecurity model, until now reserved for organizations in the United States and the United Kingdom.
- Microsoft expands its model catalog to more than 11,000 — Azure’s Foundry platform now brings together more than eleven thousand models —including those from OpenAI and Anthropic— accessible from a single point, along with Microsoft’s own models for reasoning and code. The trend: the cloud provider as a neutral store where all the labs coexist.
In the region
The MOE report does not happen in a regulatory vacuum by chance. While Colombia reaches the runoff without a rule in force requiring AI-fabricated content to be identified, other jurisdictions are tightening the calendar: the European Union will require labeling of deepfakes of public interest starting August 2, and the bulk of its AI regulation begins to apply that same day, with fines of up to 35 million euros. The asymmetry is the usual one seen from the region: the technology to fake is cheap and available today; the rule to stop it arrives late, and when it arrives, it is designed elsewhere. Colombia is, these weeks, the most exposed case of that gap.
Launches
- Gemini 3.5 Flash — Available now. Google positions it as its fast, low-cost model: around 284 tokens per second and a price of $1.50 and $9 per million input and output tokens. The more powerful Pro version is the one expected this month.
Threads we’re following
We had been following the program through which Anthropic gives restricted access to Mythos —its cybersecurity model capable of finding zero-day flaws— first to entities in the United States and the United Kingdom. Today’s chapter expands it: Japanese entities join for the first time, led by the country’s three largest banks. The logic taking shape is that frontier cybersecurity capabilities are not offered on the open market but are handed out state by state, according to agreements and trust. For Latin America, still outside that map, the question is which row of that list it wants to be in.
When an election is played out as much on social media as at the ballot box, is it enough to observe and document disinformation, or is the point of no return not even being able to know who pays to spread it?
About this entry. It is generated automatically from public sources, without human review before publication. It may contain errors of interpretation or summary; please check each story against its original source (the links lead there) before citing it or making decisions based on it.
Doble Click is written with Anthropic models.