UK unemployment climbed to 4.9% this week and the EU stripped a worker AI-literacy right, reviving an old question: who is actually counting AI's job toll? A March survey found 59% of managers overstate AI in layoffs and Oxford Economics puts real cuts at 4.5%, yet Stanford's count of a million hires never made says the damage is bigger and quieter. The Anthropic export ban enters a second week.
Three AI labour regimes diverge: Brussels weakened worker rights, Washington asserted an AI kill switch, Beijing binds employers immediately.
A March ResumeBuilder survey found 59% of managers overstate AI in layoffs; Stanford puts the real loss at a million hires never made. Two numbers, opposite errors.
A March 2026 ResumeBuilder survey found that 59% of hiring managers admit overstating AI's role in layoffs. Oxford Economics put genuine AI-driven US cuts at 55,000 out of 1.2 million in 2025; about 4.5%.
Stanford researchers found a separate, larger effect: AI may be suppressing nearly one million US hires per year. Stanford's figure points to jobs never advertised rather than to workers fired.
US job openings jumped to 7.6 million in April while actual hires fell to 5.1 million, the signature of a labour market where employers want workers they cannot find.
US job openings reached 7.6 million in April 2026, the highest in two years, while actual hires fell to 5.1 million. LinkedIn data shows firms offering AI training dropped from 35% to 26% in a year.
The 2.5-million gap signals friction, not hiring strength. Employers are rewriting job requirements to include AI skills faster than workers can acquire them. Simultaneously, they are cutting the training that would help workers qualify.
Anthropic's Fable 5 and Mythos 5 have been switched off for foreign users for eight days. The promise of restoration 'within days' has not been kept.
United States×2FranceAnthropic's Fable 5 and Mythos 5 models entered their second week of suspension after Commerce Secretary Howard Lutnick ordered foreign national access blocked on 12 June. The G7 summit at Evian, 15-17 June, failed to reverse the controls.
Axios reported the ban followed Anthropic's refusal to allow its AI for domestic surveillance and autonomous weapons. Anthropic noted that OpenAI's competing model carries the same jailbreak vulnerability and remains on sale.
The European Parliament approved the Digital Omnibus 423 to 57 on 16 June, downgrading the duty on employers to make staff understand the AI deployed on them from 'ensure' to 'support'.
The European Parliament voted 423-57 on 16 June to approve the Digital Omnibus, pushing the AI Act's employment compliance deadline to December 2027. A key employer duty to 'ensure' worker AI literacy was weakened to 'support', removing its enforceability.
The European Trade Union Confederation called this a collapse of the legal threshold. the Council has still not formally adopted the text, leaving one procedural step before the August deadline.
UK unemployment rose to 4.9% with vacancies at a five-year low of 707,000, the ONS reported on 18 June, and the agency still names no structural cause.
UK unemployment rose to 4.9% in the February-to-April quarter, with 1.76 million out of work, the Office for National Statistics reported on 18 June. Vacancies fell to 707,000, the lowest since early 2021, and real pay grew just 0.1%.
The Bank of England's AI worst-case model assumed 500,000 extra unemployed from a vacancy base above 730,000. Britain is NOW below that threshold, and official data still show nothing on how much AI is contributing.
AB 2656 would force California's public employers to give unionised staff 45 days' notice before deploying AI, and cleared the Senate Labor committee 5-0 on 17 June.
California's AB 2656 passed the Assembly 72-2 and cleared the Senate Labour committee 5-0 on 17 June. It would require public employers to give 45 days' notice before deploying AI in unionised roles. A companion bill to collect AI-impact data cleared the same committee the same day.
The bills cover only public-sector union workers, not private companies. Both still need a full Senate vote before reaching Governor Newsom.
ServiceNow cut hundreds of staff on 11 June citing 'real AI efficiencies', and no WARN Act notice appeared in any state register.
ServiceNow cut hundreds of employees on 11 June citing 'real AI efficiencies', while announcing plans to hire for AI-specific roles. No federal mass-layoff notice appeared in state filings.
The dual message, cutting existing staff while hiring AI talent, is a common restructuring pattern among enterprise software firms. The missing notice suggests either the cuts fell below the threshold or were spread across locations.
The Directors Guild board approved a four-year studio contract on 12 June; the members' ratification vote runs until 25 June, with AI footage rights at its centre.
The directors' union board approved a four-year studio deal on 12 June covering AI footage rights. A member ratification vote closes 25 June. The deal includes no ban on AI training and no residuals for AI-generated content.
The guild accepted weaker AI terms than the writers' and actors' unions won in 2023. That sets a lower precedent for creative industry AI protections going into future negotiations.
A 2026 layoff tracker counted 267 separate rounds hitting roughly 185,894 workers this year, with 56% of the companies citing AI.
A 2026 layoff tracker recorded 267 events affecting about 185,894 workers, with 56% of companies citing AI, according to TrueUp and SkillSyncer data. The figure represents roughly one corporate layoff event per calendar day.
The 56% AI-citation rate reflects what companies stated, not what research shows actually caused the cuts. Oxford Economics calculated that genuinely AI-driven cuts accounted for about 4.5% of US layoffs in 2025.
Salesforce cut 86 roles from its MuleSoft and Marketing Cloud units on 13 June, a small round from a company that has made flat headcount its declared policy.
Salesforce cut 86 roles from its MuleSoft and Marketing Cloud units on 13 June 2026. Both divisions handle integration and automation work that Salesforce's own AI platform, Agentforce, is designed to perform.
The cuts are small relative to Salesforce's 72,000-person workforce but follow CEO Marc Benioff's stated January 2026 strategy of reducing headcount as AI absorbs specialist tasks.
Upward on two tracks simultaneously. On the US-Anthropic track, the named decision point is whether Anthropic files for a federal injunction against the EAR application; if it does, the legal basis for AI export controls over SaaS is tested for the first time and the administration either settles or loses the novel authority it just established. On the EU track, the named trigger is the Council's formal adoption vote before August: if a blocking minority emerges, the December 2027 high-risk deadline defaults to the original 2025 AI Act schedule, removing the simplification concession that held the 423-57 coalition together.
AI-assisted, human-edited under the editorial responsibility of Bannermedia Ltd. Reviewed by Ed Woodcock on 20 June 2026. Editorial standards.
