Politicians love nothing more than a convenient scapegoat. When Australian politician Andrew Hastie admitted that an automated assistant drafted a tone-deaf response to a constituent discussing voluntary assisted dying, the media rushed to grab their pitchforks. Headlines shrieked about cold machines taking over human compassion, and moral panic cascaded across social media feeds.
Everyone fell for the lazy narrative. Recently making news in this space: China’s Viral AI Phenom Hides a Brutal Manufacturing Trap.
The public outcry assumes that software made a moral error. That framing is completely backwards. Technology lacks intent, conscience, or agency. Hastie owned an inbox, a communication channel, and a direct line to a human being facing the ultimate crisis. Blaming an LLM for writing "congratulations!" to a man planning to end his life is like blaming a pen for writing a death threat.
Strip away the hysteria and look at the actual workflow. A constituent emails a high-ranking politician about a deeply sensitive, tragic topic. A human staffer, or the politician himself, delegates the intake to a language model without reading the incoming message first, or accepts a generated draft blindly. The failure point sits squarely between the keyboard and the chair. More insights into this topic are covered by CNET.
For years, I have watched organizations treat automation as a magical substitute for accountability. They pump incoming correspondence into algorithms to save time, treating communication like an assembly line widget. When the inevitable trainwreck happens, they point the finger at the code. This is institutional cowardice disguised as technological critique.
Language models operate on pattern matching. Feed an unread text into an automated prompt without context parsing, and the model outputs statistical averages. If the preceding messages in a thread resemble a routine constituent acknowledgment, the algorithm defaults to standard pleasantries. It did not misread the room because it has no capacity to read a room. It processed tokens based on mathematical probability.
The real scandal here has nothing to do with artificial intelligence going rogue. The scandal is the complete automation of empathy by leaders who cannot be bothered to read their own mail.
The Myth Of Automated Empathy
We have built a culture obsessed with outsourcing emotional labor. Companies deploy chatbots to handle customer complaints about bereavement. Public figures use algorithms to draft condolences. We expect software to fake human warmth so we can optimize our schedules.
When that fake warmth misfires, we act shocked.
Consider the mechanics of the incident. Hastie's office received an email. Somewhere in the pipeline, text generation tools were hooked up to routine correspondence workflows. The expectation was efficiency. The result was a grotesque mismatch of tone.
Critics argue that we must restrict machine learning tools in sensitive domains. That solution misses the mark entirely. Banning automated drafting tools in political offices will not restore genuine human care. It will only force staffers to copy and paste generic human-written templates instead, with the exact same level of detached indifference.
The problem is not that the machine wrote the reply. The problem is that nobody stopped to read the constituent's pain.
If you are going to use advanced linguistic tools, you carry a non-negotiable burden of verification. I have seen startups blow millions scaling customer support automation only to watch brand trust evaporate because executives treated verification as optional. Leadership wants the speed without the oversight.
Technology amplifies existing human behaviors. If your office culture treats constituents as ticket numbers rather than human beings, automation will simply hyper-accelerate that alienation.
Reclaiming Operational Accountability
Fixing this requires abandoning the comforting illusion that software can feel, care, or judge context.
First, draw a hard line around high-stakes communication. Any correspondence involving mental health, crisis intervention, legal threats, or medical decisions must pass through human hands exclusively. No drafts. No automated summaries. No shortcuts.
Second, treat language models as drafting engines, never as decision-makers or communication proxies. When an algorithm generates text, it provides a raw clay shape. A human artisan must carve out the meaning, verify the context, and inject genuine situational awareness. If you lack the time to read what you send, you lack the right to send it.
Finally, stop accepting technological determinism as an excuse for poor leadership. When a politician blames an algorithm, demand their resignation not because the software failed, but because they tried to automate their core duty to listen.
The machine did exactly what it was programmed to do. The humans failed to do their jobs.
Stop policing the code. Start policing the leaders.
