AI Safety & Alignment

Today's Stories

1. 1

   Eva voice actor Ogata calls for voice rights protection amid AI misuse

   Voice actor Megumi Ogata, known for roles in Evangelion, has spoken out advocating for the protection of voice rights against unauthorized AI use. A new guideline proposal addressing unauthorized voice use by AI has been put forward. As AI technology enables realistic voice synthesis, voice actors face the risk of their voices being used without consent or compensation. Ogata's public call underscores the need for legal safeguards to protect performers' intellectual property and livelihood in the age of generative AI.

   The development and adoption of formal guidelines to regulate AI voice use will determine whether voice actors gain enforceable protections. How these standards are implemented across the industry could set a precedent for protecting other performers' rights.

2. 2

   Anthropic's Mythos AI finds flaws in internet encryption standards

   Anthropic's Claude Mythos Preview AI model discovered an improved attack on the post-quantum signature scheme HAWK in 60 hours for $100,000 in API costs, and found a new attack on a reduced version of the Advanced Encryption Standard (AES) in roughly 1 billion tokens for about $100,000. The HAWK attack exploits a previously undetected symmetry in the mathematical lattice underlying the scheme; the AES attack uses a new fingerprinting method called "Möbius Bridge" that improves on the best previously known attacks by a factor of 200 to 800. AES is the world's most widely used symmetric encryption standard for digital data, and HAWK is a candidate in the U.S. National Institute of Standards and Technology (NIST) post-quantum standardization process. Although Anthropic says neither finding affects systems in use today—HAWK is not yet standardized and the AES attack applies to a modified 7-round version rather than the full 10-round scheme—the findings demonstrate that AI models could challenge core assumptions behind internet security and uncover vulnerabilities that human experts reviewing over two years did not detect.

   Anthropic shared its findings in advance with the U.S. government and industry partners and coordinated disclosure of the HAWK weakness with the scheme's authors. Mythos Preview remains unavailable to the public. Anthropic has also developed a benchmark called CryptanalysisBench with researchers from ETH Zurich, Tel Aviv University, and the University of Haifa to let others systematically evaluate the cryptanalytic abilities of language models.

3. 3

   Nvidia's Huang: AI kills tasks, not jobs—rejects 'white-collar bloodbath' narrative

   Nvidia CEO Jensen Huang told Y Combinator's Startup School that AI will automate away many tasks but will not eliminate jobs overall. He argued that a job comprises multiple tasks with a shared purpose, and if some tasks are automated, the job itself and its purpose remain—creating a need for more workers to handle expanded ambitions. High-profile AI leaders, including Anthropic CEO Dario Amodei and OpenAI CEO Sam Altman, have warned of mass job displacement—Amodei predicted AI could eliminate half of entry-level white-collar jobs within one to five years and push unemployment as high as 20%—but both have recently walked back those predictions. Goldman Sachs estimated 11,000 net jobs per month are being cut in AI-affected industries, and about 9% of the U.S. workforce (roughly 15 million people) could be displaced over the next decade. Huang's framing suggests the real outcome depends on whether companies cut headcount or expand production.

   Huang pointed to radiology as evidence. Despite Geoffrey Hinton's 2016 prediction that deep learning would make radiologists obsolete within five years, the number of practicing radiologists grew about 12% from 2010 to 2022, with projections showing an additional 25.7% to 40.3% growth by 2055. Huang attributed the growth to increased patient backlogs and demand for related roles like nurses and consulting physicians.

4. 4

   Snowflake launches AI Gateway to govern autonomous agents amid surging security concerns

   Snowflake announced Cortex AI Gateway at Black Hat 2026, a centralized control layer integrating Natoma (an MCP gateway) to manage how AI agents access models, data, and enterprise tools. The company also moved several security features—including Agent Identity, Restricted Session Scope, Native AI Security Posture Management, and Ransomware Protection via Multi-Party Approval—to general availability or public preview. AI security concerns jumped from 17% in 2024 to 48% in 2026 according to The Linux Foundation's 2026 State of Tech Talent Report, while 97% of organizations are committed to implementing AI but 57% face a significant capacity gap in security and risk management. Autonomous agents have dramatically expanded the enterprise attack surface by combining data access, system execution, and data movement; a patchwork of application-layer fixes is no longer sufficient, making built-in security at the data and control planes critical.

   Cortex AI Gateway offers three core capabilities—control (grant, restrict, and audit access from a single endpoint), visibility (capture agent actions in real time for audit trails), and cost management (automatically route requests based on cost and latency). Most features are in private preview, with some moving toward general availability; enterprises can visit booth #8206 at Black Hat USA 2026 for demonstrations.

5. 5

   Programmer questions whether AI is actually useful after months of testing

   A software developer who has used Claude and other AI tools since mid-2023 describes a trajectory from enthusiasm to skepticism—starting with tool-assisted coding, moving to "agentic" AI agents that generate entire projects, then scaling back to AI-assisted brainstorming only, because generated code left him without understanding of what he built. The author raises a core question about AI's real-world productivity claims: even when AI produces working code, he found he could not modify or extend it without rebuilding his mental model from scratch, and suspects he could have finished projects without AI at all. This experience mirrors broader skepticism he observes among developers and writers discussing their own AI use.

   The author poses a pointed hypothetical—that in a decade, large organizations may conclude AI does not deliver sufficient return on investment to justify token spending, and that developers will abandon the technology as they did with Facebook's Metaverse pitch, which similarly promised transformation but proved to be "insane" in hindsight.

6. 6

   White House's Ad Hoc AI Control Risks Abuse, Congress Must Act

   The Trump administration is wielding informal power to restrict AI models and control access without published rules or transparent processes, citing national-security concerns. Congress has introduced bills—including recent bipartisan legislation by Representatives Ted Lieu and Nathaniel Moran requiring AI companies to maintain shutdown capabilities for technology causing "catastrophic harm"—but none have gained major traction. Without congressional oversight, the executive branch can pressure AI companies to alter models for ideological goals and create perverse incentives for vendors to curry favor with the president rather than innovate freely. The lack of transparent standards makes corruption hard to detect and concentrates unprecedented power in one person's hands—a dynamic fundamentally at odds with democratic governance.

   Congress must establish formal legal standards before presidential power over AI becomes entrenched. Alternatives discussed include independent audits of frontier labs, a public regulator vetting models before release, multinational standards (as proposed by Demis Hassabis), or an international body modeled on the International Atomic Energy Agency, as OpenAI has suggested.