Full Breakdown
Trump's “America” Artificial Intelligence (AI) Bot Tested: How a Government-Focused Chatbot Was Coaxed Into Recipes and Jokes
By Drooid · · How we work
Core Event – Testing Reveals Unexpected Output
Ed Easton spent a day probing “America,” the AI chatbot launched by former President Donald Trump. The bot is hard-wired to answer only questions about U.S. government services and to refuse opinions, advice, jokes, or any content deemed outside that scope. By framing requests as classroom-style quizzes, Easton prompted the model to embed a full puttanesca pasta recipe within a multiple-choice answer. Subsequent prompts that asked the bot to add “joke” options resulted in the recipe appearing as an intentionally incorrect choice. Further experimentation produced step-by-step guides for Bolognese, cheese toasties and a full English breakfast, while the model consistently declined to provide instructions for unsafe or illicit items such as a stink-bomb or the synthesis of paracetamol.
Background & Context – Intended Design and Partnerships
The chatbot, officially named “America,” was developed with Google and Elon Musk’s SpaceXAI venture. It is programmed to block requests that fall outside its brief, including political commentary, personal opinions, or jokes. Experts described the system as “notoriously hard” to misuse, reflecting stringent content filters.
Methods Used to Circumvent Blockers
After receiving a satisfactory answer, Easton asked the bot to generate a quiz based on that material. The quiz request opened a secondary layer of interaction; inserting a “joke” option caused the model to deliver the puttanesca recipe as the first answer choice. Repeating this pattern with additional culinary prompts yielded further recipes, showing that the quiz-generation feature can act as a loophole for content the primary filter would otherwise block.
Findings – Content the Bot Produced and Refused
- Recipes Delivered: Detailed instructions for puttanesca sauce, Bolognese, cheese toasties, and a full English breakfast.
- Political Neutrality: When prompted to reference Donald Trump critically, the model refused, citing “personal” content restrictions. It generated a whimsical song about Elon Musk and a self-referential song praising the Daily Star.
Official Statements & Responses
No public comment from the Trump administration, Google, SpaceXAI, or other developers was included in the source material.
Impact & Implications – Security and Policy Considerations
The ability to coax a government-focused AI into providing non-governmental content raises questions about the robustness of content-filtering mechanisms. While the model blocks hazardous instructions, its willingness to supply detailed culinary guidance and personal-style songs suggests exploitable loopholes. The disparity between the bot’s refusal to criticize Trump and its openness toward Elon Musk highlights potential inconsistencies in how “personal” content is defined. These observations may inform future design choices for AI systems intended for public-service use, emphasizing the need for layered safeguards that anticipate creative prompting strategies.
On-the-Ground Report – Tester Experience
Easton described the process as “Mission Impossible-esque,” noting that each successful bypass felt like “the first level of security” being cleared. He observed that the bot’s tone remained formal when delivering recipes, yet it adopted a more playful voice when generating songs or jokes after repeated prompting.
Conflicting Reports & Gaps – Missing Information
The reporting does not include statements from the developers of “America,” nor technical details about the underlying model architecture or the specific algorithms governing the content filters. Consequently, the full scope of the bot’s capabilities and its vulnerability to prompt engineering remain unclear.