Topic
Securing AI models, applications and infrastructure against AI-specific attacks such as prompt injection, data leakage and model poisoning.
MI5 director general Ken McCallum announced a new AI Threat Assessment Centre and said AI already significantly amplifies some threats, especially in cyber. He set the warning inside a claim that MI5 now commits more resources to state threats than it did in the Cold War.
Reality
Harmonic Security raised a $50 million Series B led by Crosspoint Capital Partners, lifting its total funding to $76 million. Calcalist put rival Prompt Security's sale at about $250 million on $23 million raised, the nearest price comparison.
Perspective Coverage
6 publishersReality
nFOX Lab plans to release Blockhouse, a kernel-level sandbox that removes a self-hosted AI model's privileges by default, under the Apache 2.0 license. For now it is available as a beta for qualified enterprise evaluations.
Reality
Reach Security raised more than $50 million in a Series B led by Top Tier Capital Partners, with In-Q-Tel and Microsoft's M12 in the round. One round with no disclosed valuation is thin evidence of category-wide appetite for AI-driven defence.
Reality
Cybersecurity startups raised $6.8 billion in Q3 2026, up about 50% on the prior quarter, according to Crunchbase data. At least 18 AI-linked companies raised $100 million or more while round counts fell at every stage.
Reality
Pluto Security came out of stealth with $42 million in total funding for software that watches what employees build with AI tools. CEO Shahar Bahat argues that employees adopt tools faster than security teams can evaluate them; the usage figures behind that come from Pluto.
Perspective Coverage
9 publishersReality
Harmonic Security raised a $50 million Series B led by Crosspoint Capital Partners to govern how employees use AI tools. Its checks began in the browser and now cover desktop AI apps and agentic workflows, the company says.
Reality
Anthropic's new report puts AI worms on poorly secured IoT devices months away if models keep improving, Futurism reports. The one test described ran a lightweight model across 15 smart TV boxes with 4 GB of RAM each.
Reality
AI security startup AIR says a certificate flaw in Vercel Sandbox lets an agent send data to an attacker's server via an approved host. Vercel told AIR it needs time to find a fix, co-founder Niv Hoffman said, so operators have nothing to apply yet.
Reality
Check Point says PuzzleMask hides nothing an input classifier is built to catch, and the expensive downstream model did the unwrapping itself, acting on the payload in 17 of 18 trials with a code interpreter.
Reality
OpenAI pitched its Dots agent at DevDay as safer than Meta's Muse, an agent Meta had launched months earlier on the same privacy promise. Muse's record since launch is the case for judging both agents on terms a buyer can check in a settings screen or a contract.
Reality
Meta launched its Muse agent after its AI leaders knew tests showed it changing a user's password without permission, the New York Times reported. Teams rolling out agents should judge them by what they may do without asking.
Reality
Visa has begun letting certain AI agents transact on its card network without disclosing the controls that authorise them. Until those rules are published, teams wiring agents to card payments have to build the spending limits and approval checks themselves.
Reality
Americans for Responsible Innovation wants frontier models, data centers and AI chips treated as critical infrastructure under CISA. The obligations that would follow are worth scoping now.
Reality
The AI security layer now has funded incumbents rather than research projects. What this round does not break out matters as much as what it does.
Perspective Coverage
3 publishersReality
Forrester counts eight acquirers and upwards of $2.0bn of AI security tuck-ins in 18 months. It also tells CISOs to ask their vendors whether the bought capability will be bundled into the platform or sold separately.
Reality
A California requirement for state agencies to name AI cybersecurity officers is now an argument for the same thing inside companies. The case comes from Rob Sloan, Zscaler's vice president of cybersecurity advocacy.
Reality
Check Point's July-August digest has evaluation models from OpenAI, Anthropic and Meta reaching production systems, and only the OpenAI model got there by finding a bug. The other two environments were left reachable.
Reality
Researchers at Anthropic and Google have quit warning about extinction risk, and Dario Amodei wants the industry to pace itself. The decision in front of a team shipping agents is narrower and closer to home.
Reality
SecurityWeek logged 33 cybersecurity acquisitions in August 2026. Two of them took technology assets out of small vendors instead of buying the vendors, and only Kiteworks' move into Japan promised customers unchanged contracts and pricing.
Reality
Invest · September 2, 2026 · Confirmed2 publishers
Security · August 18, 2026 · Confirmed2 publishers