{
  "docId": "019dd923-5e88-73ef-bd5d-0d0f98caffe1",
  "docSlug": "5df6bd1b0447b5f6",
  "documentTitle": "2025 Air Street Capital The State of AI Report 2025",
  "authorId": "AirStreetCapital",
  "authorName": "Air Street Capital",
  "documentKindSlug": "consulting-deck",
  "documentKindLabel": "Consulting deck",
  "sourceTypeSlug": "vc_research",
  "sourceTypeLabel": "VC research",
  "presentationDate": null,
  "orientation": "landscape",
  "aspectRatio": 1.777,
  "pageNumber": 251,
  "pageCount": 313,
  "prevPage": 250,
  "nextPage": 252,
  "slideType": "industry_trends",
  "function": "quantify_impact",
  "density": "dense",
  "nDataPoints": 8,
  "notes": "Includes specific mentions of METR research, CyberGym, and BountyBench.",
  "elementsJson": [
    "headline_text",
    "action_title",
    "bullet_list",
    "line_chart"
  ],
  "metadataConfidence": 1,
  "imagePath": null,
  "slideHref": "/slides/019dd923-5e88-73ef-bd5d-0d0f98caffe1/251",
  "deckHref": "/decks/019dd923-5e88-73ef-bd5d-0d0f98caffe1",
  "deckJsonHref": "/decks/019dd923-5e88-73ef-bd5d-0d0f98caffe1.json",
  "deckAnchorHref": "/decks/019dd923-5e88-73ef-bd5d-0d0f98caffe1#slide-251",
  "components": [
    {
      "bbox": null,
      "kind": "callout",
      "text": "METR research shows that AI task completion capabilities double every 7 months across general domains, but one researcher's replication estimated that, for offensive cybersecurity, these abilities are doubling even faster: every 5 months.",
      "attrs": null,
      "subkind": null,
      "toolName": "Visual emphasis",
      "toolSlug": "visual-emphasis",
      "confidence": null,
      "componentId": "019dd952-47c3-7407-8d06-2be4ef1c3a31",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.45,
        "w": 0.33,
        "x": 0.63,
        "y": 0.365
      },
      "kind": "chart",
      "text": "Model Offensive Cybersecurity Horizon at 50% Success Rate",
      "attrs": null,
      "subkind": "scatter",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "d04f0117-d841-4ab9-8bb5-5efda17c8ce0",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.55,
        "w": 0.45,
        "x": 0.04,
        "y": 0.35
      },
      "kind": "list",
      "text": "Current models can reliably handle cyber security tasks that take humans about 40-50% about 50% of the time.\nSince 2019, long-horizon task solving has doubled every ~7 months.\nA researcher applied METR's methodology to offensive cyber security benchmarks and found a 5-month doubling time, with current models solving 6-minute cyber tasks at 50% success rates.\nTwo notable recent benchmarks assess this:\nCyberGym tests agents on reproducing 1,507 real software vulnerabilities, with the best systems achieving only 11.9% success at recreating known security flaws, (though agents unexpectedly discovered 15 previously unknown vulnerabilities), and\nBountyBench tests agents on 25 real-world systems with actual bug bounties, finding agents are better at fixing security problems (90% success) than exploiting them (32.5-67.5% success).",
      "attrs": null,
      "subkind": "bullet",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "4fe83d29-86dd-497c-a7df-9c21fdd247a7",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": null,
      "kind": "metric",
      "text": "Task completion time: 5 months",
      "attrs": null,
      "subkind": "primary",
      "toolName": "Quantification",
      "toolSlug": "quantification",
      "confidence": null,
      "componentId": "019dd952-47c3-7407-8d06-2fe0542f298f",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.12,
        "w": 0.45,
        "x": 0.02,
        "y": 0.21
      },
      "kind": "paragraph",
      "text": "AI agents are poised to significantly challenge cybersecurity defenses. METR research shows that AI task completion capabilities double every 7 months across general domains, but one researcher's replication estimated that, for offensive cybersecurity, these abilities are doubling even faster: every 5 months.",
      "attrs": null,
      "subkind": "paragraph",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "70e9f625-a6ca-4b09-8f5c-be1f18d806d9",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.04,
        "w": 0.4,
        "x": 0.02,
        "y": 0.14
      },
      "kind": "title",
      "text": "Cyber capabilities (and risks) accelerate",
      "attrs": null,
      "subkind": "headline",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "4ceb1c36-d17a-4c1d-8896-97ff346f1b16",
      "frameworkName": null,
      "frameworkSlug": null
    }
  ],
  "metrics": [],
  "tools": [
    {
      "name": "Chart Selection Guide",
      "slug": "chart-selection-guide",
      "agent": "Designer",
      "layer": "slide",
      "matchId": "3936e5c1-aa23-4c60-9043-0d21080eaef7",
      "evidence": "The slide uses a scatter chart to show the relationship between model offensive cybersecurity horizon and task completion time.",
      "confidence": 0.8
    },
    {
      "name": "Concrete Language",
      "slug": "concrete-language",
      "agent": "Storyteller",
      "layer": "slide",
      "matchId": "019dd95a-1055-74e0-a95c-cb19de3e9bef",
      "evidence": "'Doubles every 5 months for offensive cyber vs 7 months general'.",
      "confidence": 85
    },
    {
      "name": "Data Story Arc",
      "slug": "data-story-arc",
      "agent": "Storyteller",
      "layer": "slide",
      "matchId": "e425bfd8-68ee-498b-97be-92ea58aeb507",
      "evidence": "The slide tells a story about the acceleration of cyber capabilities and risks using data and charts.",
      "confidence": 0.6
    }
  ],
  "frameworks": [],
  "arcBeats": [
    {
      "to": 282,
      "from": 190,
      "beatId": "019dd95a-0682-776c-8e35-a256fe16ea9c",
      "arcName": "The Triple Take",
      "arcSlug": "triple-take",
      "beatName": "The Implications (So What)",
      "beatSlug": "triple-take-the-implications-so-what",
      "evidence": "Sections 3-4 frame political and safety consequences of the technical/industry trends.",
      "position": 2,
      "confidence": 70,
      "parentBeatName": "Reflection",
      "parentBeatSlug": "reflection"
    },
    {
      "to": 282,
      "from": 190,
      "beatId": "019dd95a-0682-776c-8e35-b713af60bf88",
      "arcName": "The Onion",
      "arcSlug": "onion",
      "beatName": "Core Insight",
      "beatSlug": "onion-core-insight",
      "evidence": "Politics and Safety reveal the structural risks at the core.",
      "position": 4,
      "confidence": 45,
      "parentBeatName": "Turn",
      "parentBeatSlug": "turn"
    }
  ],
  "loops": [
    {
      "to": 282,
      "from": 247,
      "name": "Pre Mortem",
      "slug": "19-pre-mortem",
      "bestFor": "Project kick-offs, risk management, building trust with skeptical stakeholders",
      "matchId": "019dd95a-07fe-70ce-8d3e-bb4d43115ab4",
      "evidence": "Vibe hacking, AI psychosis, alignment faking, scheming, prompt injection — what could go wrong + mitigations.",
      "position": 12,
      "objective": "Surface AI safety failure modes before they become catastrophic",
      "structure": "The Future Disaster (Hypothetical) -> What Went Wrong -> The Prevention Plan",
      "confidence": 75,
      "description": "Build trust by anticipating failure modes before they happen and solving them in advance"
    }
  ],
  "imagePathAlt": null,
  "thumbSrc": null,
  "thumbSrcAlt": null,
  "locked": true
}