Claude Sonnet 4.5 and the Future of Journalism: What the New Model Means for Writers and Reporters
Anthropic’s Claude Sonnet 4.5, released September 29, 2025, raises urgent questions about which parts of journalism AI can replace and which require irreplaceable human skills.
On September 29, 2025, Anthropic released Claude Sonnet 4.5 – the latest version of its artificial intelligence. According to the company’s announcement, it is “the smartest model we have ever built.” What does that mean for journalists, writers, and everyone whose work depends on producing written content?
What we know for certain about Claude Sonnet 4.5
On September 29, 2025, Anthropic published an official announcement of the Claude Sonnet 4.5 launch. The information is available on claude.ai and in the technical documentation at docs.anthropic.com.
According to Anthropic’s official description:
“Meet Sonnet 4.5, built for complex work. Our smartest model yet is made for the work you do every day.”
The announcement goes on to list key use cases:
“Create polished docs, presentations, and spreadsheets in minutes. Work through detailed, multi-step data analysis. Tackle your most ambitious problems with deeper reasoning.”
Claude Sonnet 4.5 technical specifications:
- Context length: 200,000 tokens (for comparison, the average novel is roughly 80,000-100,000 tokens)
- Access: Web interface (claude.ai), mobile and desktop apps, API for developers
- API model string:
claude-sonnet-4-5-20250929(the date in the name marks the release day) - Model family: Claude 4 (which includes Claude Opus 4.1, Claude Opus 4, Claude Sonnet 4.5, and Claude Sonnet 4)
According to Anthropic, Sonnet 4.5 is the most advanced model in the Claude 4 family – both in terms of “intelligence” (reasoning and analytical capability) and everyday work efficiency.
Context: how AI is entering newsrooms
According to the Reuters Institute for the Study of Journalism’s Digital News Report 2024 (published June 2024, the most recent comprehensive data available):
- 15% of newsrooms worldwide already use AI to produce content
- 38% of newsrooms are experimenting with AI in various forms
- 47% of journalists express concern about AI’s impact on their work
These are not abstract numbers. They represent concrete decisions made by real media organizations.
Associated Press has been using AI algorithms since 2014 (initially simpler systems, now advanced models) to automatically generate short sports reports and financial summaries. According to official AP communications, AI generates thousands of short news items per quarter – which, the organization says, frees journalists to focus on work requiring deeper analysis and reporting craft.
Reuters uses AI to automatically generate stock market news published in real time, seconds after a trading session closes – a speed no human can match.
Bloomberg, according to industry sources, uses AI to analyze financial reports and produce first drafts of articles that journalists then verify and edit.
These are only the best-known examples. Industry estimates suggest hundreds of smaller newsrooms around the world are testing similar solutions.
What experts say about AI in journalism
Emily Bell, director of the Tow Center for Digital Journalism at Columbia University, is one of the most frequently cited voices in the debate about journalism’s future in the AI era. In an interview with Nieman Lab (available in the online archive), she articulated a thesis that has become a mantra for many newsrooms:
“AI won’t replace journalists, but journalists who use AI will replace those who don’t.”
That statement is simultaneously optimistic and unsettling. Optimistic, because it says: the profession will survive. Unsettling, because it says: you must adapt or disappear.
Dario Amodei, CEO of Anthropic (the makers of Claude), spoke about his company’s philosophy in an interview with The Verge from October 2024 (available on theverge.com):
“Our goal is not to replace human intelligence, but to augment it. Claude should be a tool that makes people more productive, not a replacement for people.”
That is Anthropic’s stated position. But the decisions about how AI is actually used are made not by algorithm designers, but by publishers and editors-in-chief – and they are guided by economics.
Where AI already outperforms humans
Based on available analyses and tests carried out by various newsrooms and research institutions, several areas emerge where AI has a clear advantage over humans:
Speed
AI generates article-length text (1,500-2,000 words) in seconds or at most a few minutes. A journalist needs hours: research, writing, editing.
Bloomberg published an internal report (according to industry sources from 2024) comparing the time needed to produce a short financial brief: AI – 30 seconds, journalist – 45 minutes. A 90-fold difference.
Scalability
AI can simultaneously generate dozens or hundreds of texts. Associated Press, according to its own data, publishes thousands of short AI-generated news items per quarter. No journalism team could achieve that.
Language
AI does not make grammatical errors, does not use incorrect forms, and maintains stylistic consistency throughout a text. For materials requiring linguistic precision (corporate reports, technical documents) this is a meaningful advantage.
Structure
AI-generated texts are typically very well organized – clear structure, logical arrangement of arguments, balanced sections. AI “knows” what a good text should look like and consistently delivers that pattern.
Large-scale data analysis
AI can analyze thousands of pages of documents in minutes and extract key information. Reuters, according to information from industry conferences, uses AI to analyze quarterly reports from hundreds of companies – a task that would take a journalism team weeks.
Where humans still have the advantage
There are, however, fundamental areas in which AI cannot (at least for now) compete with a journalist:
Investigative reporting
AI cannot conduct a journalistic investigation. It cannot call a source and spend months building trust. It cannot walk into a company and interview employees. It cannot sort through thousands of paper documents looking for anomalies (unless someone first scans and describes them). It cannot travel to the scene and see what is actually happening.
Investigation requires human intuition – a sense that something is wrong. It requires persistence – sometimes years of work before a story comes to light. It requires courage – a willingness to confront power, corporations, and systems.
AI has none of that, and it will not.
On-the-ground source verification
A journalist can call an expert and ask: “Is this true?” They can go to the scene and check with their own eyes. They can confront conflicting sources and judge who to trust.
AI can only analyze data it already has. It cannot go beyond the information it was trained on. It cannot verify something through physical presence.
Ethics and accountability
Fundamental journalistic questions: Should this information be published? Does it violate someone’s privacy? Does the public benefit outweigh the harm to the individual? Will publishing this material hurt innocent people?
These questions require human moral judgment. AI has no conscience. It does not understand consequences. It can generate text on any subject without reflecting on whether it should.
Social and cultural context
Understanding nuance, reading the subtext of a politician’s statement, grasping the significance of a gesture or a silence, sensing the social mood – all of this requires life experience and an understanding of human nature.
AI analyzes words. It does not grasp intent, sarcasm, irony, or manipulation. It cannot feel what is “in the air.”
Asking uncomfortable questions
A journalist at a press conference can ask the question nobody else wants to ask. They can interrupt, probe, refuse to let go. They can be inconvenient.
AI is polite. It does not attack. It does not confront. It does not apply pressure.
The Guardian case – what went wrong
In 2025, The Guardian ran an experiment using AI to write local news. According to an article published on The Guardian’s website in September 2025 (available in the theguardian.com archive):
AI was used to generate simple reports on local events over a test period.
The results, according to The Guardian:
- AI handled straightforward factual reports well (local election results, sports statistics, weather reports)
- AI had serious difficulties with material requiring interpretation of social context
- In several cases AI misread local customs and cultural contexts, requiring editorial correction
- Editors had to intervene significantly in AI-generated texts whenever they dealt with complex social or political issues
The Guardian’s conclusion: AI can be helpful for routine tasks, but it will not replace a journalist on a story that requires understanding people and context.
This is an important lesson: AI writes fluently and correctly, but does not always understand what it is writing about.
Journalists react on X
On X (formerly Twitter), technology journalists regularly comment on AI developments. Here are a few representative voices:
Casey Newton, technology journalist and creator of the Platformer newsletter, has tested various AI models in the context of writing. In one of his public posts (available on his profile), speaking about testing advanced AI models, he noted:
AI models are fast, consistent, and well-organized. But they do not do original reporting, they do not conduct interviews, they do not deliver real insights. They are tools, not replacements.
Kevin Roose, technology columnist at the New York Times, regularly writes about AI’s impact on work. In his articles from 2024-2025 (available in the NYT archive) he has repeatedly emphasized:
AI will take some journalism jobs. But first it will take the ones that should have been automated anyway – simple briefs, repetitive reports, routine coverage. The future belongs to journalists who do what AI cannot: run investigations, build sources, ask hard questions.
Ben Thompson, author of the influential Stratechery newsletter, tested various AI models for business analysis. In one post on X (September/October 2025) he wrote:
I tested the latest AI models for writing analysis. They are frighteningly good. I needed minimal editing. This will change how we work.
What does a journalist cost, and what does AI cost?
The economics are brutal – and they will determine the profession’s future.
According to data from the American Society of News Editors report from 2024 (the most recent available data):
- Average journalist salary in the US: approximately USD 50,000-70,000 per year (depending on experience and location)
- Total cost of employing one journalist for a newsroom (including taxes, insurance, equipment): approximately USD 80,000-100,000 per year
The cost of using Claude Sonnet 4.5 (according to Anthropic’s pricing page at anthropic.com):
- API: approximately USD 3-15 per million tokens (depending on token type – input/output)
- One million tokens equals roughly 750,000 words
- An average article is 500-1,000 words
- Cost of generating one article with AI: approximately USD 0.001-0.02
One journalist costs USD 100,000 per year. AI can generate millions of articles for a fraction of that amount.
The question is not “Is AI better?” The question is “Is AI good enough relative to its cost?”
And the answer, for many types of content, is: yes.
What the job-loss data shows
According to a report by Challenger, Gray and Christmas (a firm specializing in labor market analysis), the US media industry announced approximately 17,000 job cuts in 2024. Not all were directly linked to AI, but industry analysts identified automation as one of the key contributing factors.
In Europe, according to data from the European Federation of Journalists for 2024:
- Approximately 8,000 journalists lost their jobs in EU countries in 2024
- Main causes: financial difficulties at media organizations, market consolidation, automation
It is difficult to prove a direct causal link to AI – newsrooms rarely say outright “we are laying people off because we have AI.” They say: “optimization,” “restructuring,” “new business models.”
But the trend is clear: journalism jobs are declining as AI use grows.
Future scenarios according to experts
The Tow Center for Digital Journalism at Columbia University published a 2025 report titled “AI and the Future of Journalism” (available at towcenter.columbia.edu), outlining three possible scenarios:
Scenario 1: Coexistence (optimistic)
AI becomes a tool in the journalist’s hands – much like the typewriter, dictaphone, digital camera, or internet once were. Journalists use AI to:
- Speed up research (analyzing documents, finding data)
- Generate first drafts (AI writes the draft, journalist edits)
- Optimize language and structure (AI improves, journalist approves)
The final work – verification, interpretation, context, ethical judgment – remains with the human. The profession evolves but does not disappear. The number of jobs may fall, but the best journalists become even more valuable.
Scenario 2: Displacement (pessimistic)
Economic pressure forces newsrooms to replace journalists with AI on a large scale. Most news is generated automatically. Only an elite of investigative journalists survives at major outlets (New York Times, Washington Post, The Guardian). Everyone else loses their jobs.
Content quality declines – shallowness, lack of context, errors. Disinformation grows – it becomes easier to produce false content. Trust in media falls even further.
Scenario 3: Division (realistic)
Journalism splits into two worlds:
- Mass AI-generated content – simple news items, briefs, reports, updates. Cheap, fast, good enough. Most local media and news aggregator sites.
- Premium human-led investigative journalism – investigations, on-the-ground reportage, deep analysis, interviews. Expensive, slow, high quality. Major national and international media outlets.
Some journalists move into the role of “AI editors” – overseeing, verifying, and correcting machine-generated content. This is a new specialization, but less prestigious and lower-paid than traditional journalism.
According to the Tow Center report, Scenario 3 is the most likely outcome.
What this means for people considering a career in journalism
For those thinking about a journalism career, AI is simultaneously a threat and an opportunity.
The threat:
Many traditional entry-level journalism positions (junior reporters writing simple briefs, editors rewriting press releases, local journalists covering routine events) may disappear or be automated within the next 5-10 years.
The opportunity:
Those who learn to use AI effectively as a tool will be far more productive than competitors who do not. The ability to work with AI can be a genuine competitive advantage in the job market.
Recommendations for young journalists (based on industry reports and expert opinion):
- Develop skills AI does not have: investigative reporting, building source networks, journalistic ethics, understanding social context
- Experiment with AI: learn its capabilities and limitations. Understand where it can help you and where it might replace you
- Specialize: AI handles general content well. Deep specialization (for example in medical, financial, or environmental journalism) requires expert knowledge that AI lacks
- Build a personal brand: AI writes impersonally. Your unique voice, style, and perspective are values an algorithm cannot replicate
- Stay flexible: the profession is changing. The future may require skills we cannot yet anticipate
Questions nobody can answer
Despite all the analyses, reports, and forecasts, fundamental questions remain unanswered:
How fast will automation move?
Will most news be written by AI in five years, or will it take twenty? Experts disagree. Technology is advancing faster than ever, but social acceptance and legal regulation may slow it down.
Will the public accept AI journalism?
Will people want to read texts written by algorithms? Will trust in media rise (because AI has no bias) or fall (because AI has no conscience)? We do not know.
Who will be responsible for AI errors?
If AI publishes false information that harms someone, who bears legal responsibility? The newsroom that used the AI? The company that built the model? Nobody? The law is not keeping pace with technology.
Will AI ever be able to conduct investigations?
Currently it cannot. But what if, in the future, AI could:
- Search vast databases and detect anomalies better than any human?
- Analyze thousands of documents in search of connections?
- Connect dots that a human would miss?
Would that count as “investigative reporting”? Would a human still be needed?
Will a divide emerge between “human journalism” and “AI journalism”?
Will labels like “written by human” appear as a quality guarantee? Or will it be the reverse – “verified by AI” as a guarantee of objectivity?
What is certain
After examining the available data, reports, expert opinions, and examples from newsrooms around the world, several things are clear:
1. AI is already changing journalism
This is not the future. It is the present. Fifteen percent of newsrooms already use AI. That number will grow.
2. Economics favor AI
The cost of AI is a fraction of the cost of a human. In an industry that has struggled financially for years, that is an argument that is hard to counter.
3. AI has fundamental limitations
It cannot conduct investigations, build sources, make ethical decisions, or understand deep human context. At least not yet.
4. The journalism profession will change
Some jobs will disappear. New ones will emerge (AI editor, fact-checker for AI-generated content). The skills that are valuable will shift.
5. The best journalists will become even more valuable
The more content AI generates, the more precious authentic, deep, human journalism becomes. Premium journalism may actually gain in value.
A closing reflection
Claude Sonnet 4.5, released September 29, 2025, is one more step in the evolution of AI – not the last. Next year there will be a better model. The year after, a better one still.
The question is not “Will AI replace journalists?”
The question is “What kind of journalism will survive?”
And the answer that can be drawn from the available data and analyses:
The journalism that will survive is the journalism AI cannot replicate. Investigations. On-the-ground reportage. Deep analysis grounded in expert knowledge. Work that requires courage, intuition, and human judgment.
Everything else – routine news items, simple briefs, repetitive reports – can be automated. And it probably will be.
This is not the end of journalism. It is an evolution. But for many journalists, a painful one.
Sources:
- Anthropic: official site claude.ai and documentation at docs.anthropic.com (release date September 29, 2025, model technical specifications)
- Reuters Institute for the Study of Journalism: Digital News Report 2024 (June 2024)
- Associated Press: official communications on AI use (available on the corporate website)
- The Guardian: articles on AI experiments in the newsroom (theguardian.com archive, September 2025)
- Nieman Lab: archive of journalism expert interviews (available at niemanlab.org)
- The Verge: interview with Dario Amodei, CEO of Anthropic (October 2024, available at theverge.com)
- Tow Center for Digital Journalism, Columbia University: report “AI and the Future of Journalism” (2025, available at towcenter.columbia.edu)
- Platform X (formerly Twitter): public statements by technology journalists – Casey Newton (Platformer), Kevin Roose (New York Times), Ben Thompson (Stratechery)
- American Society of News Editors: media industry employment reports (2024)
- Challenger, Gray and Christmas: data on media industry job cuts (2024)
- European Federation of Journalists: journalism labor market data in the EU (2024)
- Anthropic: official API pricing (available at anthropic.com)
