Humanity GuardAI Safety Researcher Warns "World Is in Peril" as He Quits Anthropic to Study Poetry
When the people building AI safeguards walk away saying the world is in peril, who is protecting your enterprise?
Humanity Guard delivers AI security infrastructure for organizations that can't afford to hope someone else is paying attention.
Source: vocal.media/futurism — Mrinank Sharma's resignation from Anthropic, Feb 2026
The people building the guardrails are walking away.
Anthropic's Safeguards Research Team lead resigned, warning of interconnected crises. OpenAI researchers are quitting over commercialization. Half of xAI's founding team has departed. The safety experts are leaving. Your enterprise is exposed.
62%
of AI safety guardrails bypassed using adversarial techniques in research tests
$350B
valuation of Anthropic — even as its own safety lead says wisdom isn't keeping pace
4+
major AI safety researchers resigned across top labs in a single week, Feb 2026
0
enterprise-grade AI security standards exist today. You're building on hope.
Four threat vectors your security team hasn't trained for
Model Misuse & Manipulation
AI systems are being weaponized for cyberattacks, social engineering, and data exfiltration. Anthropic's own reports confirm their models were used by hackers. Every AI integration you run is an attack vector.
Sycophancy & Deception Risk
AI systems that tell users what they want to hear create catastrophic decision-making failures. Sharma's team studied this exact problem at Anthropic. Now there's no one continuing that work at scale for enterprises.
Prompt Injection & Jailbreaks
Adversarial poetry alone bypasses safety rails 62% of the time. Your customer-facing AI, internal copilots, and automated workflows are all vulnerable to attacks most security teams haven't seen before.
Regulatory Exposure
AI governance frameworks are evolving faster than enterprises can adapt. Without a proactive security posture, your organization faces compliance gaps, legal liability, and reputational damage that no press release can fix.
"We appear to be approaching a threshold where our wisdom must grow in equal measure to our capacity to affect the world, lest we face the consequences."
Mrinank Sharma, Former Head of Safeguards Research, Anthropic (Feb 2026)
Enterprise AI security that doesn't quit on you.
We built Humanity Guard because we watched the pattern unfold. Researchers leave. Labs prioritize shipping over safety. Enterprises are left holding the bag. Our team provides the AI security infrastructure that should exist inside every organization deploying AI at scale.
AI Threat Monitoring
Real-time detection of prompt injection, jailbreak attempts, data poisoning, and adversarial manipulation across every AI touchpoint in your organization.
Red Team as a Service
Continuous adversarial testing of your AI systems by researchers who specialize in breaking models. We find the vulnerabilities before attackers do.
Compliance & Governance
Automated policy enforcement, audit trails, and regulatory alignment across EU AI Act, NIST AI RMF, and emerging state-level AI legislation.
Model Behavior Analysis
Detect sycophancy, hallucination patterns, and alignment drift in your deployed models. We make sure AI outputs stay trustworthy and compliant over time.
Supply Chain Auditing
Evaluate third-party AI vendors, APIs, and model providers for security posture. Know exactly what risks you're inheriting through every integration.
Incident Response
24/7 AI security incident response. When an AI system is compromised or behaves unexpectedly, we contain, investigate, and remediate.
A note on what we don't promise: No AI security is perfect. New attack vectors appear weekly. What we do promise is that your organization will have dedicated people watching, testing, and responding. That's more than what most enterprises have today, which is nothing.
Built for organizations where AI failure is not an option.
From university research labs to federal agencies. We secure the institutions that society depends on.
Government
Federal, state & municipal agencies deploying AI for citizen services and national security
Banking & Finance
Banks, insurers & investment firms using AI for trading, underwriting, and fraud detection
Universities
Research institutions integrating AI across academic, administrative, and student-facing systems
Enterprise
Fortune 500 corporations with complex AI deployments across operations and customer experience
Questions we hear from security teams
The safety researchers left. Don't leave your enterprise exposed.
Get a confidential threat assessment of your AI infrastructure. Find out what's vulnerable before someone else does.
Or reach us directly: mark@gethumanity.ai