It's a lot, and now it's even more: Freelance creatives are bearing ever more cost and risk (Midjourney) Companies stand to capture enormous upside from AI: access to more specialised capability, faster production and less need to maintain every skill in-house. But a growing share of the cost and uncertainty required to produce that capability is being pushed in the opposite direction, onto the individual workers supplying it.Freelancers and contractors are expected to fund rapidly changing technology, continuously retrain themselves, absorb experimentation that may never generate revenue and, more often, accept legal responsibility for risks embedded in models they did not build and cannot inspect. AI may be making companies more efficient partly by making individual creative workers carry more of the downside.Part of a bigger pictureAI is extending an existing labour-market shift that has transferred more risk from companies to individual workers. Australia already has more than a million independent contractors and millions more people working casually, on fixed-term contracts or through labour hire. These arrangements give businesses flexibility, but also shift more uncertainty onto the worker. A permanent employer absorbs periods of lower utilisation, paid leave, illness and at least some of the cost of maintaining a worker’s capabilities.An independent worker carries much more of that risk personally. Generative AI is now extending that transfer into areas the labour market has barely begun to price. For independent creatives, staying commercially relevant now means arriving not only with the right craft skills, but with an entire technological stack already in place.Enjoying Mumbrella? Sign up for our free daily newsletter.ADVERTISEMENT A quick poll of our AI Artist network tells me that a fairly ordinary professional setup costs around $1,000 a month. That figure is still climbing as platforms move more capability into higher-tier subscriptions, compute charges and credit-based plans.Independent workers have, of course, always paid for some of their own tools. Photographers buy cameras, designers buy computers and editors pay for software.The difference with generative AI is the pace and uncertainty of the investment. A camera can be amortised over years and hundreds of jobs. An AI workflow can become second-best within months, sometimes weeks.The author Vinne Schifferstein VidalModels change, pricing changes and yesterday’s advantage can quickly become today’s baseline. The artist therefore carries not only the cost of the technology but the risk that an investment in a particular workflow will cease to be commercially valuable before it has paid for itself.The same is true of learning. Historically, companies had an incentive to invest in the skills of permanent employees because those capabilities sat inside the business and could be used over time. That investment still happens. But as companies buy more specialist capability from freelancers, contractors and external partners, they have less reason to fund training in those externally sourced skills.The expectation is that independent workers arrive already current, with the cost of staying current absorbed into their own business. AI makes this particularly stark because expertise depreciates unusually fast. Knowing one tool is less valuable than being able to continuously test, evaluate and integrate new ones.For many independent creatives, that learning happens between jobs, at night or through speculative work that may never generate revenue. The buyer gets access to the finished expertise without necessarily paying for the process that created it.One-sided dealThis is efficient for the buyer, but the economics look different from the other side of the transaction. The individual is being asked to operate more like a small production company, absorbing training, infrastructure, experimentation and technological obsolescence, while still being purchased largely as labour. Those costs are rarely itemised. They disappear into a day rate and become the freelancer’s overhead. What looks like labour-market flexibility can therefore become entrepreneurial risk without entrepreneurial pricing.The most revealing extension of this logic is now appearing in contracts. Brands are understandably nervous about copyright, provenance, likeness, ownership and training data, so legal and procurement teams are trying to contract that risk away.We see it directly in agreements sent to us as a small production business, and the artists in our wider network see it in the contracts passed on to them. The obligations can extend well beyond copyright: suppliers may be asked to stand behind legal and ethical compliance, prevent client data being used for model training, manage risks such as bias and hallucination, document how AI has been used and take responsibilities for whether outputs infringe third-party rights.That liability can move from brand to agency, agency to production company and production company to AI artist, until an individual is being asked to stand behind risks arising partly from a foundation model they did not build, cannot inspect and barely influence.No precedent for where risk belongsWho bears the legal risk for this generated picture of concerned freelancers considering who bears the risk of generated pictures? (Midjourney)What makes this particularly striking is that the courts have barely begun to establish where those risks ultimately belong. The major generative-AI copyright cases are still being fought much further upstream. Disney, Universal and Warner Bros have sued Midjourney over allegations that its models reproduce protected characters and record labels have pursued Suno and Udio.The legal system is still wrestling with the responsibility of the companies that build the technology, yet commercial contracts can already ask an individual artist using that technology to warrant that everything coming out of it is safe. The market is allocating liability before the courts have finished allocating responsibility.The irony is that some technology companies already recognise that this risk belongs higher up the chain: Adobe, for example, introduced intellectual-property indemnification for eligible enterprise customers using Firefly, offering protection against certain third-party copyright claims connected with generated outputs.If a model provider can decide that AI-related intellectual-property risk is sufficiently connected to its technology to warrant indemnifying a large enterprise customer, why should an individual artist automatically be expected to absorb the same category of risk further downstream?Yet liability can still end up with the person who has the least control over how the model was trained and the least capacity to survive a serious legal claim. Risk does not always flow to the party best equipped to manage it. Often it flows to the party least able to refuse it.Suddenly, independence doesn’t look so goodFor many talented AI artists, independence is precisely the attraction. They want the freedom to choose projects, clients and creative problems. But that freedom now comes with a formidable list of obligations: fund your own downtime, train yourself continuously, finance the tools, absorb failed experimentation and indemnify the organisations buying the work against risks embedded in technology you do not control.Advertising is debating productivity, headcount and capability. The economic question is who gets the upside and who carries the downside. If companies can access ever more sophisticated external capability while pushing more of the cost, uncertainty and liability required to sustain it onto individuals, then AI is doing more than changing how creative work is made. It is extending a much older transfer of risk from institutions to workers. Companies may become more resilient precisely because the individuals they depend on are being asked to absorb more of the cost, uncertainty and risk. Topics Adobe Disney Firefly midjourney Suno Universal Vinne Schifferstein Vidal Warner Bros Have your say Or comment anonymously Your comment will be marked as unverified Name Comment Δ Latest Burrowes and Simons go head-to-head in podcast showdown Amazon upfront: Youth focus, industry engagement and Australian radio added to inventory pool On the hook: AI pushes legal risk and big costs onto individuals Tom Steinfort and Sylvia Jeffreys to host Today show Sarah Abo to leave Today and Nine