What the AI press predicted.
Each day after the agent run, we extract falsifiable predictions from the top stories — specific forecasts about what’s going to happen. Readers vote on whether they think each one will play out. When the horizon arrives, we search fresh news and resolve the call. The numbers below are the receipts.
Currently open — vote or wait
Malicious AI agents using hijacked models will be linked to at least one new autonomous cyber-attack by 2026-06-30.
Anthropic's technical workforce will be larger in absolute terms within 12 months than at the time of this report.
By 2027-10-01, a 10,000-employee company will be able to cut about 2,000 roles using AI agents, as Eric Schmidt predicted.
Storm-0940 will be reported targeting additional organizations with ARTEX-enabled attacks beyond those disclosed as of 2026-10-21, within 90 days.
Microsoft will publish follow-up detection guidance or updated ARTEX-related indicators for defenders before 2027-04-21.
The FDA will publish draft guidance on generative AI in medical devices within the next few months.
Leading the Future will spend millions to elect US candidates opposing strict AI rules in the 2026 midterms.
The Association of Human Mathematicians will still hold at least 1,000 pledged members six months from now.
OpenAI will issue another bulk release of AI-generated mathematics papers within 12 months.
Sub-Saharan African countries will strengthen implementation of existing data protection and cybersecurity rules by end-2028, per World Bank recommendation.
The U.S. federal government will not enact new AI liability legislation in 2026.
Google will publish a migration guide for using TorchTPU to move PyTorch apps to TPUs by 2026-12-20.
Google will make its internally deployed Argon model available to external developers through Google Cloud within six months.
Google will publicly claim its frontier models materially accelerated development of a successor model before 2027-01-01.
New victim organizations beyond the 100+ already reported will be identified in Cairn-abuse attacks within 90 days.
At least one additional case of Cairn being used for malicious cyber activity will be documented by 2027-10-19.
Embedded third-party AI evaluators will not be accepted as a substitute for active government AI regulation by 2027-10-19.
AI evaluations will shift toward testing the agent harness and surrounding scaffolding rather than the model alone by 2027-10-19.
Australia will announce or introduce copyright changes allowing AI training on copyrighted music by 2027-10-19.
Australia will remain the only country among those approached to water down AI copyright rules over the next 12 months.
Users will continue to download open AI models, remove refusal behaviors, and re-upload them publicly within the next 6 months.
AI model developers will not consistently implement existing misuse safeguards within the next 12 months.
The U.S. federal government will not require government-overseen pre-release testing of AI models before 2027-01-01.
By 2027-05-07, the EU AI Act will still not require public disclosure of serious incidents reported by systemic-risk GPAI providers.
Disney will complete hiring of at least 60 Character.AI employees, the majority of its workforce, by 2027-04-16.
Australia will not introduce new copyright legislation specifically for AI national security before 2027.
Persistent skill shortages in technology, healthcare and education will remain across OECD advanced economies through 2027-05-09.
CyberSN's next Cybersecurity Jobs Report will show AI Security Engineer demand up at least 50% year-over-year.
US cybersecurity job postings will stay below 300,000 in the January-September 2027 reporting period.
Databricks will occupy 120,000 square feet across five floors at 5 Bryant Park by 2027-04-18.