Pharma Bloodbath V: Last Blood. The AI displacement panic edition. What thirteen years watching Pharma's technology waves has taught me about survival

Welcome back to Pharma Bloodbath, the newsletter that speaks about uncomfortable truths without the usual marketing BS and always tries to help you cope with dry humour and silly German phrases.

Parts I through IV covered the 2024 bloodbath, the structural forces behind it, the cautious optimism of early 2026, and the patent cliff that is currently eating BMS's lunch with both hands and asking for a side of fries. Part V was supposed to be about AI inside pharmaceutical recruitment processes specifically. The wider AI displacement panic has, however, gatecrashed the discourse with such enthusiasm in 2026 that ignoring it would feel professionally negligent. Manchmal kommt es eben dicker als geplant. (Sometimes things turn out thicker than planned. Like a Stallone sequel.)

The subtitle is Last Blood , which is, yes, a deliberate reference to Rambo: Last Blood (2019), the fifth instalment in a franchise widely understood to have peaked around instalment two and to have spent every film since loudly insisting otherwise. The closing scene of Last Blood shows an ageing John Rambo, having survived the Vietnamese jungle, the Burmese army, the Afghan mujahideen and a small Mexican cartel, rocking quietly on a porch in Arizona. The audience response was, to put it diplomatically, mixed. The metaphor, however, is almost embarrassingly on the nose. An ageing super-soldier, declared obsolete by every younger generation that has come of age since First Blood (1982), shows up one last time to do exactly the thing he has always done, kills a small army doing it, then sits down for a cup of coffee. Das Original altert. Aber es zieht sich nicht freiwillig zurück. (The original ages. But it does not retire voluntarily. It needs to be told. Repeatedly. And then it kills you anyway.)

The professional pharma workforce in 2026 is being told, with rising volume and falling subtlety, that it is structurally obsolete. This is the same message Rambo received in 1982, 1985, 1988, 2008 and 2019. The franchise grossed approximately $839 million across the five films despite each instalment being declared the last. Make of that what you will.

First, the numbers. Because panic without data is just astrology with footnotes and a slightly more aggressive thumbnail.

The figures circulating in your feed, sorted by approximate decibel level and inversely correlated with how often anyone has read past the headline:

The ifo Institut, June 2025: 27.1% of German companies expect artificial intelligence to lead to job cuts within the next five years. In industry specifically, 37.3%. Affected companies forecast an average 8% headcount reduction. Only 5.2% expect AI to create additional jobs. ¹ The figure is real. It is also, on careful reading, an expectation about a five-year horizon, gathered via a survey, from managers who five years ago were forecasting blockchain-based supply chains and the end of the office. Da liegt der Hund begraben. (That is where the dog is buried.) Corporate forecasts about technology and their relationship to subsequent reality have, historically, occupied a similar emotional register to weather predictions in a coastal British town.

Mercer Global Talent Trends 2026: 40% of employees worldwide now fear losing their job to AI , up from 28% in 2024. ² Twelve percentage points in eighteen months. Whether the actual job risk has moved by twelve points or whether the fear has, is the kind of distinction a competent epidemiologist would draw before recommending a treatment. The discourse, regrettably, is being shaped by neither competent epidemiologists nor any other species of inconvenient empiricist.

Gartner, Q1 2025: only 26% of job applicants trust AI to evaluate them fairly. 25% say they trust an employer less the moment AI shows up in the screening process. 39% of candidates cheerfully admit to using AI to write their own applications. ³ Which means, on a strict reading of the evidence, that the AI in the applicant tracking system is currently being politely overpolished by the AI in the cover letter, while the human being on neither side of the exchange is entirely sure what just happened or who is supposed to be evaluating whom. Ein digitales Pingpong-Spiel ohne Spieler, aber mit Servicegebühren auf beiden Seiten des Netzes. (A digital ping-pong match without players, but with service fees on both sides of the net.)

April 15, 2026: UKG, the HR software company that sells the very technology other companies use to manage their own restructuring, laid off 950 of its own employees and named the cause in writing: "rapidly evolving market shifts, including changes in technology driven by AI." ⁴ Six per cent of the workforce. Approximately a year after the previous round of 2,200, which had also been blamed, with the consistency of a man who has discovered exactly one excuse and intends to ride it until retirement, on AI. The symbolic charge of this case cannot reasonably be overstated. The company that helps you fire people fired its own people, blamed the software it sells, and presumably used the software it sells to do it. Operation gelungen, Patient tot. (Operation a success. The patient died. But the operating room was very efficient about it.)

May 2025: Anthropic CEO Dario Amodei predicts AI could eliminate half of all entry-level white-collar jobs within five years and push unemployment to 10 to 20%. ⁵ The warning is delivered with the gravitas of a man who has read both the data and his own product roadmap. May 2026, twelve months later: the same Dario Amodei, sitting onstage with JPMorgan's Jamie Dimon, pivots to the Jevons Paradox, a nineteenth-century theory about coal consumption which, applied to labour, suggests AI will probably create more work than it destroys, actually. ⁶ The arc from "white-collar bloodbath" to "actually, William Stanley Jevons had a rather underappreciated insight about steam engines" in the space of twelve months is, to put it generously, an interesting personal journey. Manche nennen das Lernen. Andere nennen das Marketing. Manche nennen es: man hatte zwischenzeitlich einen Pentagon-Vertrag zu beschützen und der ursprüngliche Tonfall war strategisch suboptimal. (Some call it learning. Others call it marketing. Some call it: there was a Pentagon contract to protect in the meantime and the original messaging had become strategically suboptimal.) The CEO of the AI company that builds the products that would do the displacement, oscillating between warning the world about his own products and reassuring the world about his own products within a single fiscal year, is the kind of internal contradiction the franchise needs.

This is the data. None of it is fictional. None of it tells the whole story either. Ein Schelm, wer Böses dabei denkt. (A rogue is he who thinks ill of it.)

Four technology waves. Zero apocalypses. One slightly humiliating pattern.

Late 1990s, the genomics revolution: medicinal chemists were supposed to be obsolete by 2010. A surprising number of trade publications committed this prediction to print. The medicinal chemists were not obsolete by 2010. The medicinal chemists who could not read a sequence file were obsolete by 2008. There is a small but career-defining difference between those two sentences, and the people who learned it the hard way are still finding new ways to phrase it on LinkedIn.

Mid-2000s, clinical operations outsourcing to India: the end of the Western clinical career was confidently predicted, often by the same people who had predicted the end of medicinal chemistry. It did not arrive. What arrived was the surgical end of the Western CRA whose competitive advantage consisted of geographical proximity to the sponsor and the ability to drive to the investigator site without complaining about traffic. The CRAs who learned to manage globally distributed teams across four time zones continued to be paid quite reasonably. The ones who refused continued to insist, into the early 2010s, that the work could not be done remotely. Totgesagte leben länger. Bürobesitzer auch. (Those declared dead live longer. Office holders too.)

The 2010s, digital health and electronic data capture: traditional CRA roles were declared finished by a fresh wave of consultants who had not been present for the previous wave of consultants. They were not finished. They were trimmed at the routine end and thickened at the judgement end, which is the same thing every technology wave has done since the introduction of the typewriter and which, with a really humbling regularity, every cycle of business journalism has failed to remember.

2020, the pandemic remote shift: the office career was over, pharma hiring was about to democratise across continents, geography was dead. As we are all observing in 2026, that particular obituary has been quietly retracted, with employers walking back to hybrid and on-site arrangements at the speed of a man who has just remembered he left the stove on, his keys at his ex-wife's house, and a perfectly good office lease on a twenty-year covenant. Aus dem Regen in die Traufe (out of the rain and straight under the gutter) turns out to apply to corporate real estate strategy with the same enthusiasm it applies to most other adult decisions.

The pattern, after sixteen years inside the machine, is so consistent it has stopped being amusing and started being a useful planning tool. Each wave predicts a generalised extinction. Each wave delivers a surgical contraction. Each wave eliminates the routine version of a role. Each wave expands the judgement-heavy version of the same role. Each wave quietly humiliates the cohort that refused, on principle, to engage with the new tools while there was still time and the certifications were still cheap. The professionals who actually lost their jobs were almost never the ones literally replaced by the technology. They were, with a consistency that should worry anyone reading this, the ones who treated the new technology as someone else's department, right up to the day on which someone else's department turned out, after a brief restructuring meeting, to be theirs. Wer sich verweigert, wird nicht gefeuert. Wer sich verweigert, wird vergessen. Und vergessen zu werden ist in der Pharmaindustrie eine spezielle Form von Tod, die niemand auf LinkedIn ankündigt. (Whoever refuses to engage does not get fired. Whoever refuses to engage gets forgotten. And being forgotten, in pharma, is a specific form of death that nobody announces on LinkedIn.)

The 2026 AI wave differs from its four predecessors in three specific ways: faster capability curve, wider functional reach, higher cognitive content of the work being touched. It does not differ in structural shape. The discourse is louder than the substance. The substance is surgical rather than apocalyptic. The people who will be fine in five years are, in nearly every case I am observing, already doing one or two of the things this article is about to list. The people who will not be fine are, with equal predictability, currently posting that they refuse on principle to engage with ChatGPT because it is unethical, while their less principled colleagues use it to write the cover letter that gets them the next role. Holzauge, sei wachsam. (Wooden eye, stay alert. And consider blinking occasionally.)

The DACH picture, and a quick glance over the garden fence

Germany. The ifo 27.1% figure is real, frequently cited, and rarely contextualised. The IAB study published in November 2025 places a useful sibling figure next to it: approximately 1.6 million German jobs structurally affected by AI by 2035, with total employment broadly stable. Around 110,000 net new jobs in IT and information services. Around 120,000 net losses in corporate services. ⁷ A redistribution, not a vanishing. The press release that translates "1.6 million jobs redistributed" into "AI Doom Approaches Germany" has, somewhere, a Schlagzeilenredakteur (headline editor) who deserves a quiet word from their compliance team. For DACH pharma specifically, the exposure is concentrated in commercial back-office functions, document drafting, and routine pharmacovigilance case processing. Regulatory affairs, clinical data science, biostatistics, GMP-anchored manufacturing and quality assurance roles remain in structural demand, with hiring managers I speak with on a weekly basis still struggling to fill open positions and resorting to recruiters who, on the morally weaker days, charge them for the privilege. The April 2026 ifo skills-shortage figure hit a five-year low of 22.7%, ⁸ but the average disguises persistent acute shortages in exactly the technical functions pharma needs most. Eine Schwalbe macht noch keinen Sommer, und ein bisschen Personalverfügbarkeit macht noch keine bequeme Stellenbesetzung. (One swallow does not make a summer, and a little personnel availability does not make a comfortable hire.)

United States. Loud and frequent citation of AI as the cause of restructuring. UKG, Block, Salesforce, Meta, Snap, Amazon, Microsoft, Google all named AI in Q1 2026 press releases with the unanimity of a barbershop quartet. Volkswagen's 35,000 European cuts have been partly framed as automation-driven realignment, although the more honest framing involves an electric vehicle strategy that may require its own newsletter, possibly its own genre of newsletter. The US benefits from a light regulatory framework and fast cycles. The US suffers from a light regulatory framework and fast cycles. SignalFire reports graduate hiring at the largest tech firms down roughly 50% from pre-pandemic levels. ¹¹ For European pharma professionals contemplating the US, the headline opportunities are in manufacturing reshoring rather than office-based commercial work, which is a meaningful detail for anyone who has not previously worked in a bioprocessing plant in Indiana, was perhaps not planning to start now, and does not own appropriate footwear.

United Kingdom. Lighter regulatory framework than the EU, faster AI rollout across CRO and consultancy roles, earlier substitution patterns at the routine end. For DACH candidates contemplating relocation, this is worth a moment's thought before the application form opens. The British have already had Brexit. They are not, on balance, in a hurry to manufacture their next regret.

China and India. China's domestic biotech venture market is in its third post-2022 contraction year. AI in CDMO drug discovery is being deployed at a tempo that EU regulators would describe, after a strong coffee, as ambitious. India remains the global CRO and shared-services hub, and AI there is overwhelmingly augmenting headcount rather than replacing it, because the business case still rests on cost-effective scale rather than cost-effective elimination. For multilingual DACH professionals open to relocation, India is an underdiscussed option that no LinkedIn influencer is going to recommend, because it inconveniently fails to fit the doom narrative on which their personal brand depends.

The EU AI Act. The headline date is 2 August 2026 , when high-risk obligations for AI used in recruitment, candidate selection, performance evaluation, task allocation and termination decisions become enforceable. Bias audits, technical documentation, human oversight, EU database registration, six-month log retention as the floor. ⁹ The Digital Omnibus package, currently mid-trilogue (second round 28 April 2026, third scheduled 13 May 2026), may defer parts of this to December 2027 if adopted in time. ¹⁰ At the time of writing, it has not been adopted. Brussels is, in the politest possible terms, taking its time. Until adoption, the August 2026 deadline remains live, your works council is already entitled to be informed and consulted before deployment, and the maximum fines, designed to make GDPR look like a friendly parking ticket left under a windscreen wiper, start at €15 million or 3% of global turnover. Jetzt geht's um die Wurst. (Now it is about the sausage. Which, in German, means: now it gets serious. The translation does not quite capture the menace.)

Five strategies that survive contact with reality, hiring managers, and your own self-deception

A note before this section: the fundamentals from Bloodbath I through IV (update your LinkedIn, activate your network, get specific about your CV, target CROs, build AI literacy) all still apply. If you have not done those, please close this tab, do them, and come back. This section assumes the fundamentals are handled. If they are not, keine Schande, aber auch keine Selbsttäuschung (no shame, but no self-deception either, and the second one is harder than the first).

1. For juniors, interns, entry-level candidates, and the visa-sponsored international cohort. Audit your routine-output share with brutal honesty. If 80% of what you currently do could be done by a competent language model in a useful afternoon, your differentiation problem is structural rather than cosmetic, and no amount of polishing your CV is going to fix it. Build documented evidence, before you apply, that you can do the 20% the model cannot: judgement under regulatory ambiguity, stakeholder management with an investigator site that is two enrolments behind plan, the explanation of a GMP deviation to a non-specialist auditor who has decided today is the day. The 2026 graduate hiring decline in pharma commercial functions is real. It is also not uniform. Junior roles in clinical operations, GMP quality, regulatory affairs and pharmacovigilance remain accessible to candidates who can demonstrate the judgement layer, even at entry level. The Chancenkarte visa pathway, incidentally, treats EU AI Act compliance specialisations as priority skill profiles, which is one of the few useful policy decisions to emerge from Berlin in recent memory and possibly the only one that has not been immediately walked back.

2. For mid-career professionals. Get formally certified in one specific AI tool used in your function. Not "AI literacy", which in 2026 carries roughly the differentiating power that "computer skills" carried in 1998 and "good with email" carried in 2003. A named pharmacovigilance signal-detection platform. A named regulatory submission authoring assistant. A named clinical operations system. Most vendors run certification programmes you can complete in three to six weeks, which is less time than it takes most people to complete an Asana migration. The certificate is not the prize. The certificate is documented evidence, on your CV and your LinkedIn, of self-directed investment in staying current, which is, in a hiring market that is filtering hard for adaptability, the single most reliable signal I am observing across mid-career DACH searches right now. The alternative is putting "AI enthusiast" on your profile and praying that nobody asks a follow-up question in the interview. Wer rastet, der rostet, und Rost ist auf einem Lebenslauf besonders schwer wegzupolieren, besonders wenn das Foto auch noch von 2009 ist. (Who rests, rusts, and rust is particularly hard to polish off a CV, especially when the profile photo is also from 2009.)

3. For executives and senior leaders, up to C-level. Build judgement-heavy visibility outside your current employer's letterhead. A published article, a conference talk, a working group seat, a non-executive position somewhere unglamorous. The hiring conversations for senior pharma roles in 2026 are increasingly happening before a recruiter ever calls you, in rooms you are not in, between people who already know each other. If your only professional presence is your internal reputation at one company, then from the external market's perspective you are, regrettably, a rumour. Rumours do not get headhunted. Documented operators with public reasoning trails do. Wer im Keller arbeitet, wird im Keller vergessen. Und der Keller ist immer kälter, als man denkt. (Whoever works in the basement gets forgotten in the basement. And the basement is always colder than one thinks.)

4. For candidates 50 and above, including those who have been told one too many times that they are "overqualified", which is the corporate euphemism for "we already have a 32-year-old who will accept less". Reframe your accumulated institutional memory as risk avoidance, not as length of service. A 28-year-old hiring manager looking at your CV is doing arithmetic on cost and adaptability, whether or not they would admit it under oath, and the arithmetic is rarely flattering. The correction, the one that actually works in interviews, is to demonstrate what your experience cost the organisation to acquire and what specific avoidable disasters your presence prevents. Three major regulatory submissions, two M&A integrations, one product recall: these are not bullet points on a CV. They are insurance premiums the organisation has already paid for once and would strongly prefer not to pay for a second time. Erfahrung lässt sich nicht in vier Wochen onboarden, und der Algorithmus hat noch nie eine FDA-Inspektion überlebt. Auch kein Werkstudent. (Experience cannot be onboarded in four weeks. The algorithm has yet to survive its first FDA inspection. Neither has a working student.) John Rambo, across all five films, never lost because he was too old. He lost, when he lost, because the script demanded it. There is, in 2026, no script. There is only a hiring manager with a job to fill, a regulatory deadline closing in, a board breathing down their neck, and a CFO who is about to discover what an enforcement action costs.

5. For everybody, irrespective of seniority. The EU AI Act is, for DACH pharma, the most under-priced career opportunity of the year and one of the few stories where the regulatory burden creates more jobs than the technology destroys. The 2 August 2026 deadline is generating concrete, hireable demand for documentation, governance, bias-testing, human-oversight and compliance specialists across HR tech, pharma HR, pharma IT, and the entire CRO sector that supports all of them. If you have any regulatory, quality, GxP or compliance background, this is the single most demand-positive specialism to emerge from the AI displacement story. The professionals most worried about being replaced by AI are, with an irony that ought to be illegal, exactly the demographic the new regulatory regime requires in order to function. Manchmal frisst die Schlange den eigenen Schwanz. Manchmal wird der Schwanz dafür extra gut bezahlt. Und manchmal ist die Schlange auch nur die Beraterin, die EU-Compliance-Audits zu Tagessätzen verkauft, die ein Klinikdirektor mit Tränen in den Augen unterschreibt. (Sometimes the snake eats its own tail. Sometimes the tail is paid extra well for it. And sometimes the snake is really just the consultant selling EU compliance audits at day rates that make a clinical director sign with tears in his eyes.)

For the recruiters reading this. Three things that actually work, none of which involve buzzword bingo.

This newsletter is read by both sides of the table. A short, calibrated section for the hiring side, because the corporate influencer industry has already produced enough hollow content on this topic to fill a small landfill and is, at time of writing, working hard on a second.

First, disclose in the job advert where AI is used in your screening process. The 26% trust figure from Gartner is not a marketing problem to be solved with another round of employer branding copy and a stock photo of a diverse team laughing at a laptop. It is structural mistrust caused by candidates not knowing whether a person or a model is reading their application, and what that model is optimising for. The evidence is consistent: disclosure closes the gap measurably and, as a useful side effect, materially reduces candidate fraud, because fraudulent applicants prefer opaque processes for the same reason burglars prefer dark streets. Transparency is, as it usually is, cheaper than the alternative.

Second, where the role permits, move to skills-first hiring criteria with structured assessment. The evidence on bias reduction, particularly against the 50+ candidate pool that most DACH pharma organisations are quietly under-utilising while complaining publicly about skills shortages, is strong. This is not a diversity points exercise. It is a market arbitrage opportunity, of the kind that closes the moment your competitors notice. An experienced regulatory affairs lead with eighteen years of submission experience is, in the current market, both available and, frankly, mispriced.

Third, start your EU AI Act compliance work now. The 2 August 2026 deadline is months away even if the Digital Omnibus passes and shifts part of it to December 2027. The consultation, bias-testing, documentation and works-council cycles cannot be done at the last minute, regardless of how many consultants insist over a presentation deck that they can. Vendors are already overbooked. Your competitors who started in 2025 are six months ahead of you. Wer zu spät kommt, den bestraft das Leben, und in diesem Fall auch die nationale Marktaufsichtsbehörde, die GDPR-Bußgelder im Vergleich wie Falschparken erscheinen lässt. (Those who arrive too late are punished by life, and in this case also by the national market surveillance authority, which makes GDPR fines look like parking tickets by comparison.)

The Last Blood closing reflection. Or: why the franchise refuses to die, and why your career probably will not either.

Rambo: Last Blood was, by general critical consensus, not the strongest film in the franchise. It was, however, the film in which the ageing protagonist made it unambiguously clear that he was not going to be retired by external pressure. He was going to retire on his own porch, in his own time, on his own terms, and possibly after killing one more small army for old times' sake.

The 2026 AI displacement panic is, in shape, the fifth instalment in a franchise of "this technology will end your career" predictions that has been running, by my count, since approximately 1997. Each instalment promised mass extinction. Each delivered something narrower, more specific, and considerably less photogenic. Each rewarded the professionals who got their hands dirty with the new tools before they had to, and quietly humiliated the ones who did not. The 2026 instalment will, in my opinion, follow the same arc, even if it is faster in tempo, wider in reach, and immeasurably louder in marketing.

The professionals who will be fine in five years are, almost without exception, already doing one or two of the five strategies above. The professionals who will not be fine are the ones currently writing 800-word LinkedIn posts about why they refuse, on principle, to use any large language model, ending with the words "Thoughts?" and three hashtags, while their less principled colleagues quietly use ChatGPT to draft the cover letter that gets them the next role. Die Hoffnung stirbt zuletzt, aber das Marketingbudget der KI-Industrie überlebt sie locker. (Hope dies last, but the AI industry's marketing budget will easily outlive it.)

You read to the end. That puts you in the small minority who do. Now do something with it that does not involve a comment along the lines of "Great post Andreas" followed by three emojis. Specifically, do not include the emojis. We have collectively been through enough.

Prost. Und bleibt standhaft. (Cheers. And stay resilient.) Bloodbath VI is already in development. The topic, if you can stand any more of this, will be the specific failure modes of AI inside pharma applicant tracking systems, a subject so rich in unintentional comedy that it is essentially writing itself, which is, in this newsletter, an unsettling thought.

Your turn

This newsletter only works when readers do the harder thing and answer in the comments. A like is appreciated. A specific answer is more useful, especially for the next reader who lands here looking for a real perspective rather than another thread of motivational poster captions and stock photography.

• Which of the five strategies fits where you actually are, and what is the specific obstacle stopping you from acting on it? "I am busy" does not count.

• Recruiters: what is the most honest sentence you can write about how your organisation is using AI in screening, that you cannot put in the official job advert? Anonymity acceptable.

• For readers 50 and above: what reframing of your experience has actually worked in a recent interview, and which reframing has failed embarrassingly enough to be worth a war story?

• Have you, like Dario Amodei, quietly updated your view on AI displacement in the last twelve months? In which direction, and what changed your mind? Please show your working.

A short reminder for anyone who reached the end and wants something more structured than a newsletter article: www.morethancareer.de hosts the full overview of my career coaching (job search strategy, interview preparation, salary negotiation training with the actual scripts), reverse recruitment, and CV and LinkedIn rebranding. No bootcamps. No mailing list ransom. Just the work.

Sources

All footnotes are clickable and link directly to the original source.

Sources & References

[1] ifo Institut, June 2025: A Quarter of Companies in Germany Expect Job Cuts Due to Artificial Intelligence – https://www.ifo.de/en/press-release/2025-06-05/quarter-companies-germany-expect-job-cuts-due-artificial-intelligence

[2] Mercer Global Talent Trends 2026, reported via TheStreet and CNBC, May 2026 (40% fear AI-related job loss, up from 28% in 2024) – https://www.thestreet.com/technology/anthropic-ceo-makes-shocking-admission-about-ai

[3] Gartner Survey, July 2025: Just 26% of Job Applicants Trust AI Will Fairly Evaluate Them – https://www.gartner.com/en/newsroom/press-releases/2025-07-31-gartner-survey-shows-just-26-percent-of-job-applicants-trust-ai-will-fairly-evaluate-them

[4] HR Executive / Canadian HR Reporter / Sun Sentinel, April 2026: UKG Cuts 950 Jobs in Latest Round of Restructuring – https://hrexecutive.com/ukg-cuts-950-jobs-in-latest-round-of-restructuring/

[5] Axios, May 2025: Dario Amodei Warns AI Could Eliminate Half of Entry-Level White-Collar Jobs – https://www.axios.com/2025/05/28/ai-jobs-white-collar-unemployment-anthropic

[6] Fortune, May 2026: Dario Amodei Pivots to the Jevons Paradox at Anthropic Financial Services Briefing – https://fortune.com/2026/05/05/dario-amodei-jevons-paradox-will-ai-wipe-out-white-collar-jobs/

[7] Allwork.Space / IAB, December 2025: AI Anxiety Grows in Germany (1.6 million jobs structurally affected, total stable) – https://allwork.space/2025/12/ai-anxiety-grows-in-germany-as-one-in-six-fear-job-loss/

[8] ifo Employment Barometer, April 2026: Skills Shortage at Five-Year Low (22.7%) – https://www.ifo.de/en/survey/ifo-employment-barometer

[9] EU AI Act, Annex III Category 4 and Article 26: High-Risk Obligations for Employment-Related AI Systems, Effective 2 August 2026 – https://artificialintelligenceact.eu/what-the-act-means-for-staffing-businesses/

[10] DLA Piper GENIE, May 2026: The Digital AI Omnibus Trilogue Status (Second 28 April 2026, Third 13 May 2026) – https://knowledge.dlapiper.com/dlapiperknowledge/globalemploymentlatestdevelopments/2026/The-Digital-AI-Omnibus-Proposed-deferral-of-high-risk-AI-obligations-under-the-AI-Act

[11] TheStreet / SignalFire / CNBC, May 2026: Graduate Hiring Decline and AI-Cited US Layoffs in 2025-2026 – https://www.thestreet.com/technology/anthropic-ceo-makes-shocking-admission-about-ai